From 9ef24b77680943f5cccab19dc2094a47ae167d7d Mon Sep 17 00:00:00 2001
From: yexiaozhou <yexiaozhou@dp.tech>
Date: Wed, 1 Apr 2026 00:32:34 +0800
Subject: [PATCH] =?UTF-8?q?feat(layout=5Foptimizer):=20DE=20optimizer=20V2?=
 =?UTF-8?q?=20=E2=80=94=20custom=20loop,=20graduated=20hard=20constraints,?=
 =?UTF-8?q?=20broad=20phase?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Replace scipy differential_evolution with custom DE loop for per-device
crossover, circular θ wrapping, and configurable mutation strategy
(currenttobest1bin default, best1bin as turbo mode).

Key improvements:
- Graduate ALL hard constraints during DE (proportional penalty instead of
  flat inf), giving DE smooth gradient for reachability, min_spacing, etc.
  Binary inf preserved for final pass/fail reporting.
- 2-axis sweep-and-prune AABB broad phase for collision pair pruning
- Multi-seed injection from multiple seeder presets + Gaussian variants
- snap_theta_safe: collision-check after angle snapping, revert on violation
- Weight normalization (100 distance / 60 angle / 5× hard multiplier)
- Constraint priority field (critical/high/normal/low → weight multiplier)
  with LLM intent interpreter setting priority per constraint type
- Final success field now checks user hard constraints in binary mode
- arm_slider added to mock checker reach table (1.07m)

Tests: 202 passed, 24 new tests added (optimizer 7, constraints 6, broad_phase 11)

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
---
 unilabos/layout_optimizer/broad_phase.py      |  66 +++++
 unilabos/layout_optimizer/constraints.py      | 115 +++++---
 .../layout_optimizer/intent_interpreter.py    |  47 ++++
 unilabos/layout_optimizer/mock_checkers.py    |   1 +
 unilabos/layout_optimizer/models.py           |   2 +
 unilabos/layout_optimizer/optimizer.py        | 263 +++++++++++++++---
 unilabos/layout_optimizer/server.py           |  22 +-
 .../tests/test_broad_phase.py                 | 241 ++++++++++++++++
 .../tests/test_bugfixes_v2.py                 |   9 +-
 .../tests/test_constraints.py                 | 157 ++++++++++-
 .../tests/test_e2e_pcr_pipeline.py            |   8 +-
 .../layout_optimizer/tests/test_optimizer.py  | 224 ++++++++++++++-
 12 files changed, 1072 insertions(+), 83 deletions(-)
 create mode 100644 unilabos/layout_optimizer/broad_phase.py
 create mode 100644 unilabos/layout_optimizer/tests/test_broad_phase.py

diff --git a/unilabos/layout_optimizer/broad_phase.py b/unilabos/layout_optimizer/broad_phase.py
new file mode 100644
index 00000000..ca433a38
--- /dev/null
+++ b/unilabos/layout_optimizer/broad_phase.py
@@ -0,0 +1,66 @@
+"""2 轴 sweep-and-prune 宽相碰撞检测。
+
+对每个设备计算旋转后的 AABB，先沿 x 轴排序并剪枝，
+再用 y 轴交叠过滤。返回候选碰撞对（索引对列表），
+供后续 OBB SAT 精确检测使用。
+"""
+
+from __future__ import annotations
+
+from .models import Device, Placement
+
+
+def sweep_and_prune_pairs(
+    devices: list[Device],
+    placements: list[Placement],
+) -> list[tuple[int, int]]:
+    """2 轴 sweep-and-prune，返回 AABB 交叠的索引对。
+
+    Args:
+        devices: 设备列表，与 placements 一一对应。
+        placements: 布局位姿列表。
+
+    Returns:
+        候选碰撞对列表，每个元素为 (i, j)，
+        i < j，索引对应 placements 原始顺序。
+    """
+    n = len(devices)
+    if n < 2:
+        return []
+
+    # --- 计算每个设备旋转后的 AABB ---
+    aabbs: list[tuple[float, float, float, float]] = []
+    for dev, pl in zip(devices, placements):
+        hw, hd = pl.rotated_bbox(dev)
+        aabbs.append((pl.x - hw, pl.x + hw, pl.y - hd, pl.y + hd))
+
+    # --- 按 xmin 排序，保留原始索引映射 ---
+    sorted_indices = sorted(range(n), key=lambda k: aabbs[k][0])
+
+    # --- 扫描 x 轴，y 轴过滤 ---
+    candidates: list[tuple[int, int]] = []
+    for si in range(len(sorted_indices)):
+        i = sorted_indices[si]
+        x_min_i, x_max_i, y_min_i, y_max_i = aabbs[i]
+        for sj in range(si + 1, len(sorted_indices)):
+            j = sorted_indices[sj]
+            x_min_j, _x_max_j, y_min_j, y_max_j = aabbs[j]
+            # 由于按 xmin 排序，x_min_j >= x_min_i
+            if x_min_j > x_max_i:
+                break  # 后续设备 xmin 更大，不可能与 i 在 x 轴交叠
+            # x 轴交叠确认，检查 y 轴
+            if y_min_i <= y_max_j and y_min_j <= y_max_i:
+                # 保证输出 (min_idx, max_idx) 方便去重和测试
+                pair = (min(i, j), max(i, j))
+                candidates.append(pair)
+
+    return candidates
+
+
+def broad_phase_device_pairs(
+    devices: list[Device],
+    placements: list[Placement],
+) -> list[tuple[str, str]]:
+    """返回候选碰撞对的 device_id 字符串元组列表。"""
+    index_pairs = sweep_and_prune_pairs(devices, placements)
+    return [(placements[i].device_id, placements[j].device_id) for i, j in index_pairs]
diff --git a/unilabos/layout_optimizer/constraints.py b/unilabos/layout_optimizer/constraints.py
index d23d0f66..f726428a 100644
--- a/unilabos/layout_optimizer/constraints.py
+++ b/unilabos/layout_optimizer/constraints.py
@@ -9,6 +9,7 @@ from __future__ import annotations
 import math
 from typing import TYPE_CHECKING
 
+from .broad_phase import sweep_and_prune_pairs
 from .models import Constraint, Device, Lab, Placement
 from .obb import (
     nearest_point_on_obb,
@@ -21,6 +22,21 @@ from .obb import (
 if TYPE_CHECKING:
     from .interfaces import CollisionChecker, ReachabilityChecker
 
+# 归一化默认权重 — 1cm距离违规 ≈ 5°角度违规 的惩罚量级
+DEFAULT_WEIGHT_DISTANCE: float = 100.0   # 1cm → penalty 1.0
+DEFAULT_WEIGHT_ANGLE: float = 60.0       # 5° → penalty ~1.0
+
+# 硬约束graduated模式下的惩罚倍数
+HARD_MULTIPLIER: float = 5.0
+
+# 优先级等级对应的权重乘数
+PRIORITY_MULTIPLIERS: dict[str, float] = {
+    "critical": 5.0,
+    "high": 2.0,
+    "normal": 1.0,
+    "low": 0.5,
+}
+
 
 def evaluate_constraints(
     devices: list[Device],
@@ -29,6 +45,8 @@ def evaluate_constraints(
     constraints: list[Constraint],
     collision_checker: CollisionChecker,
     reachability_checker: ReachabilityChecker | None = None,
+    *,
+    graduated: bool = True,
 ) -> float:
     """统一评估所有约束，返回总 cost。
 
@@ -39,9 +57,10 @@ def evaluate_constraints(
         constraints: 约束规则列表
         collision_checker: 碰撞检测实例
         reachability_checker: 可达性检测实例（可选）
+        graduated: True=比例惩罚（DE优化用），False=二值inf（最终pass/fail用）
 
     Returns:
-        总 cost。硬约束违反返回 inf，否则为软约束 penalty 之和。
+        总 cost。硬约束违反在非graduated模式返回 inf，否则为加权 penalty 之和。
     """
     device_map = {d.id: d for d in devices}
     placement_map = {p.device_id: p for p in placements}
@@ -50,7 +69,8 @@ def evaluate_constraints(
 
     for c in constraints:
         cost = _evaluate_single(
-            c, device_map, placement_map, lab, collision_checker, reachability_checker
+            c, device_map, placement_map, lab, collision_checker, reachability_checker,
+            graduated=graduated,
         )
         if math.isinf(cost):
             return math.inf
@@ -66,8 +86,8 @@ def evaluate_default_hard_constraints(
     collision_checker: CollisionChecker,
     *,
     graduated: bool = True,
-    collision_weight: float = 1000.0,
-    boundary_weight: float = 1000.0,
+    collision_weight: float = DEFAULT_WEIGHT_DISTANCE * HARD_MULTIPLIER,   # 500
+    boundary_weight: float = DEFAULT_WEIGHT_DISTANCE * HARD_MULTIPLIER,    # 500
 ) -> float:
     """评估默认硬约束（碰撞 + 边界），无需显式声明约束列表。
 
@@ -86,18 +106,17 @@ def evaluate_default_hard_constraints(
     device_map = {d.id: d for d in devices}
     cost = 0.0
 
-    # Graduated collision penalty: sum of penetration depths
-    n = len(placements)
-    for i in range(n):
-        for j in range(i + 1, n):
-            di, dj = device_map[placements[i].device_id], device_map[placements[j].device_id]
-            ci = obb_corners(placements[i].x, placements[i].y,
-                             di.bbox[0], di.bbox[1], placements[i].theta)
-            cj = obb_corners(placements[j].x, placements[j].y,
-                             dj.bbox[0], dj.bbox[1], placements[j].theta)
-            depth = obb_penetration_depth(ci, cj)
-            if depth > 0:
-                cost += collision_weight * depth
+    # Graduated collision penalty: 2 轴 sweep-and-prune 宽相 + OBB SAT 精确检测
+    candidate_pairs = sweep_and_prune_pairs(devices, placements)
+    for i, j in candidate_pairs:
+        di, dj = device_map[placements[i].device_id], device_map[placements[j].device_id]
+        ci = obb_corners(placements[i].x, placements[i].y,
+                         di.bbox[0], di.bbox[1], placements[i].theta)
+        cj = obb_corners(placements[j].x, placements[j].y,
+                         dj.bbox[0], dj.bbox[1], placements[j].theta)
+        depth = obb_penetration_depth(ci, cj)
+        if depth > 0:
+            cost += collision_weight * depth
 
     # Graduated boundary penalty: sum of overshoot distances (rotation-aware)
     for p in placements:
@@ -142,17 +161,31 @@ def _evaluate_single(
     lab: Lab,
     collision_checker: CollisionChecker,
     reachability_checker: ReachabilityChecker | None,
+    *,
+    graduated: bool = True,
 ) -> float:
-    """评估单条约束规则。"""
+    """评估单条约束规则。
+
+    graduated=True 时硬约束返回比例惩罚（DE用），
+    graduated=False 时硬约束返回 inf（最终 pass/fail）。
+    """
     rule = constraint.rule_name
     params = constraint.params
     is_hard = constraint.type == "hard"
 
+    # 根据优先级等级计算有效权重
+    effective_weight = constraint.weight
+    if constraint.priority and constraint.priority in PRIORITY_MULTIPLIERS:
+        effective_weight *= PRIORITY_MULTIPLIERS[constraint.priority]
+
     if rule == "no_collision":
         checker_placements = _to_checker_format_from_maps(device_map, placement_map)
         collisions = collision_checker.check(checker_placements)
         if collisions:
-            return math.inf if is_hard else constraint.weight * len(collisions)
+            if is_hard and not graduated:
+                return math.inf
+            w = effective_weight * (HARD_MULTIPLIER if is_hard else 1.0)
+            return w * len(collisions)
         return 0.0
 
     if rule == "within_bounds":
@@ -162,7 +195,10 @@ def _evaluate_single(
                 checker_placements, lab.width, lab.depth
             )
             if oob:
-                return math.inf if is_hard else constraint.weight * len(oob)
+                if is_hard and not graduated:
+                    return math.inf
+                w = effective_weight * (HARD_MULTIPLIER if is_hard else 1.0)
+                return w * len(oob)
         return 0.0
 
     if rule == "distance_less_than":
@@ -177,7 +213,10 @@ def _evaluate_single(
         else:
             dist = _device_distance_center(pa, pb) or 0.0
         if dist > max_dist:
-            return math.inf if is_hard else constraint.weight * (dist - max_dist)
+            if is_hard and not graduated:
+                return math.inf
+            w = effective_weight * (HARD_MULTIPLIER if is_hard else 1.0)
+            return w * (dist - max_dist)
         return 0.0
 
     if rule == "distance_greater_than":
@@ -192,7 +231,10 @@ def _evaluate_single(
         else:
             dist = _device_distance_center(pa, pb) or 0.0
         if dist < min_dist:
-            return math.inf if is_hard else constraint.weight * (min_dist - dist)
+            if is_hard and not graduated:
+                return math.inf
+            w = effective_weight * (HARD_MULTIPLIER if is_hard else 1.0)
+            return w * (min_dist - dist)
         return 0.0
 
     if rule == "minimize_distance":
@@ -205,7 +247,7 @@ def _evaluate_single(
             dist = _device_distance_obb(da, pa, db, pb)
         else:
             dist = _device_distance_center(pa, pb) or 0.0
-        return constraint.weight * dist
+        return effective_weight * dist
 
     if rule == "maximize_distance":
         a_id, b_id = params["device_a"], params["device_b"]
@@ -218,11 +260,12 @@ def _evaluate_single(
         else:
             dist = _device_distance_center(pa, pb) or 0.0
         max_possible = math.sqrt(lab.width**2 + lab.depth**2)
-        return constraint.weight * (max_possible - dist)
+        return effective_weight * (max_possible - dist)
 
     if rule == "min_spacing":
         min_gap = params.get("min_gap", 0.0)
         all_placements = list(placement_map.values())
+        total_penalty = 0.0
         for i in range(len(all_placements)):
             for j in range(i + 1, len(all_placements)):
                 pi, pj = all_placements[i], all_placements[j]
@@ -233,9 +276,12 @@ def _evaluate_single(
                 else:
                     dist = _device_distance_center(pi, pj) or 0.0
                 if dist < min_gap:
-                    if is_hard:
-                        return math.inf
-                    return constraint.weight * (min_gap - dist)
+                    total_penalty += (min_gap - dist)
+        if total_penalty > 0:
+            if is_hard and not graduated:
+                return math.inf
+            w = effective_weight * (HARD_MULTIPLIER if is_hard else 1.0)
+            return w * total_penalty
         return 0.0
 
     if rule == "reachability":
@@ -268,18 +314,19 @@ def _evaluate_single(
         target_point = {"x": target_p.x, "y": target_p.y, "z": 0.0}
         target_point["_obb_dist"] = dist
         if not reachability_checker.is_reachable(arm_id, arm_pose, target_point):
-            if is_hard:
+            if is_hard and not graduated:
                 return math.inf
             # Graduated: penalty proportional to overshoot
             max_reach = reachability_checker.arm_reach.get(arm_id, 2.0)
             overshoot = max(0.0, dist - max_reach)
-            return constraint.weight * overshoot * 10.0
+            w = effective_weight * (HARD_MULTIPLIER if is_hard else 1.0)
+            return w * overshoot * 10.0
 
         # Line-of-sight penalty: penalize if any other device OBB blocks
         # the path from opening to arm
         los_cost = _line_of_sight_penalty(
             arm_id, arm_p, target_device_id, target_p,
-            device_map, placement_map, constraint.weight,
+            device_map, placement_map, effective_weight,
         )
         return los_cost
 
@@ -288,8 +335,10 @@ def _evaluate_single(
             (1 - math.cos(4 * p.theta)) / 2 for p in placement_map.values()
         )
         if is_hard:
-            return math.inf if alignment_cost > 1e-6 else 0.0
-        return constraint.weight * alignment_cost
+            if not graduated:
+                return math.inf if alignment_cost > 1e-6 else 0.0
+            return HARD_MULTIPLIER * effective_weight * alignment_cost
+        return effective_weight * alignment_cost
 
     if rule == "prefer_seeder_orientation":
         target_thetas = params.get("target_thetas", {})
@@ -301,7 +350,7 @@ def _evaluate_single(
             # Circular distance: (1 - cos(diff)) / 2 gives 0..1 range
             diff = p.theta - target
             cost += (1 - math.cos(diff)) / 2
-        return constraint.weight * cost
+        return effective_weight * cost
 
     if rule == "prefer_orientation_mode":
         mode = params.get("mode", "outward")
@@ -319,7 +368,7 @@ def _evaluate_single(
                 continue
             diff = p.theta - target
             cost += (1 - math.cos(diff)) / 2
-        return constraint.weight * cost
+        return effective_weight * cost
 
     # 未知约束类型，忽略
     return 0.0
diff --git a/unilabos/layout_optimizer/intent_interpreter.py b/unilabos/layout_optimizer/intent_interpreter.py
index f009d8ae..0f6e2a50 100644
--- a/unilabos/layout_optimizer/intent_interpreter.py
+++ b/unilabos/layout_optimizer/intent_interpreter.py
@@ -41,6 +41,7 @@ def _handle_reachable_by(intent: Intent, result: InterpretResult) -> None:
             type="hard",
             rule_name="reachability",
             params={"arm_id": arm, "target_device_id": target},
+            priority="critical",
         )
         result.constraints.append(c)
         generated.append({"type": c.type, "rule_name": c.rule_name, "params": c.params, "weight": c.weight})
@@ -64,6 +65,8 @@ def _handle_close_together(intent: Intent, result: InterpretResult) -> None:
         return
 
     weight = _PRIORITY_WEIGHTS.get(priority, _DEFAULT_WEIGHT)
+    # 映射 intent priority 到 constraint priority 等级
+    constraint_priority = "high" if priority == "high" else "normal"
     generated: list[dict] = []
     for dev_a, dev_b in itertools.combinations(devices, 2):
         c = Constraint(
@@ -71,6 +74,7 @@ def _handle_close_together(intent: Intent, result: InterpretResult) -> None:
             rule_name="minimize_distance",
             params={"device_a": dev_a, "device_b": dev_b},
             weight=weight,
+            priority=constraint_priority,
         )
         result.constraints.append(c)
         generated.append({"type": c.type, "rule_name": c.rule_name, "params": c.params, "weight": c.weight})
@@ -94,6 +98,8 @@ def _handle_far_apart(intent: Intent, result: InterpretResult) -> None:
         return
 
     weight = _PRIORITY_WEIGHTS.get(priority, _DEFAULT_WEIGHT)
+    # 映射 intent priority 到 constraint priority 等级
+    constraint_priority = "high" if priority == "high" else "normal"
     generated: list[dict] = []
     for dev_a, dev_b in itertools.combinations(devices, 2):
         c = Constraint(
@@ -101,6 +107,7 @@ def _handle_far_apart(intent: Intent, result: InterpretResult) -> None:
             rule_name="maximize_distance",
             params={"device_a": dev_a, "device_b": dev_b},
             weight=weight,
+            priority=constraint_priority,
         )
         result.constraints.append(c)
         generated.append({"type": c.type, "rule_name": c.rule_name, "params": c.params, "weight": c.weight})
@@ -131,6 +138,7 @@ def _handle_max_distance(intent: Intent, result: InterpretResult) -> None:
         type="hard",
         rule_name="distance_less_than",
         params={"device_a": device_a, "device_b": device_b, "distance": distance},
+        priority="normal",
     )
     result.constraints.append(c)
 
@@ -160,6 +168,7 @@ def _handle_min_distance(intent: Intent, result: InterpretResult) -> None:
         type="hard",
         rule_name="distance_greater_than",
         params={"device_a": device_a, "device_b": device_b, "distance": distance},
+        priority="normal",
     )
     result.constraints.append(c)
 
@@ -180,6 +189,7 @@ def _handle_min_spacing(intent: Intent, result: InterpretResult) -> None:
         type="hard",
         rule_name="min_spacing",
         params={"min_gap": min_gap},
+        priority="high",
     )
     result.constraints.append(c)
 
@@ -198,6 +208,7 @@ def _handle_face_outward(intent: Intent, result: InterpretResult) -> None:
         type="soft",
         rule_name="prefer_orientation_mode",
         params={"mode": "outward"},
+        priority="low",
     )
     result.constraints.append(c)
 
@@ -216,6 +227,7 @@ def _handle_face_inward(intent: Intent, result: InterpretResult) -> None:
         type="soft",
         rule_name="prefer_orientation_mode",
         params={"mode": "inward"},
+        priority="low",
     )
     result.constraints.append(c)
 
@@ -234,6 +246,7 @@ def _handle_align_cardinal(intent: Intent, result: InterpretResult) -> None:
         type="soft",
         rule_name="prefer_aligned",
         params={},
+        priority="low",
     )
     result.constraints.append(c)
 
@@ -246,6 +259,39 @@ def _handle_align_cardinal(intent: Intent, result: InterpretResult) -> None:
     })
 
 
+def _handle_keep_adjacent(intent: Intent, result: InterpretResult) -> None:
+    """keep_adjacent：两个设备保持相邻（同 close_together 逻辑，支持 priority 映射）。"""
+    devices: list[str] = intent.params.get("devices", [])
+    priority: str = intent.params.get("priority", "medium")
+
+    if len(devices) < 2:
+        result.errors.append(f"keep_adjacent: 参数 'devices' 至少需要 2 个设备，当前 {len(devices)} 个")
+        return
+
+    weight = _PRIORITY_WEIGHTS.get(priority, _DEFAULT_WEIGHT)
+    # 映射 intent priority 到 constraint priority 等级
+    constraint_priority = "high" if priority == "high" else "normal"
+    generated: list[dict] = []
+    for dev_a, dev_b in itertools.combinations(devices, 2):
+        c = Constraint(
+            type="soft",
+            rule_name="minimize_distance",
+            params={"device_a": dev_a, "device_b": dev_b},
+            weight=weight,
+            priority=constraint_priority,
+        )
+        result.constraints.append(c)
+        generated.append({"type": c.type, "rule_name": c.rule_name, "params": c.params, "weight": c.weight})
+
+    result.translations.append({
+        "source_intent": intent.intent,
+        "source_description": intent.description,
+        "source_params": intent.params,
+        "generated_constraints": generated,
+        "explanation": f"设备组 {devices} 应保持相邻（优先级: {priority}）",
+    })
+
+
 def _handle_workflow_hint(intent: Intent, result: InterpretResult) -> None:
     """workflow_hint：工作流顺序暗示，相邻步骤设备靠近。"""
     workflow: str = intent.params.get("workflow", "")
@@ -263,6 +309,7 @@ def _handle_workflow_hint(intent: Intent, result: InterpretResult) -> None:
             type="soft",
             rule_name="minimize_distance",
             params={"device_a": dev_a, "device_b": dev_b},
+            priority="normal",
         )
         result.constraints.append(c)
         generated.append({"type": c.type, "rule_name": c.rule_name, "params": c.params, "weight": c.weight})
diff --git a/unilabos/layout_optimizer/mock_checkers.py b/unilabos/layout_optimizer/mock_checkers.py
index 2d5ecdb9..d677a40c 100644
--- a/unilabos/layout_optimizer/mock_checkers.py
+++ b/unilabos/layout_optimizer/mock_checkers.py
@@ -78,6 +78,7 @@ class MockReachabilityChecker:
         "elite_cs66": 0.914,
         "elite_cs612": 1.304,
         "elite_cs620": 1.800,
+        "arm_slider": 1.07,   # 线性导轨臂：body 2.14m × 0.35m，reach ≈ half length
     }
 
     # 未知型号回退臂展：realistic default for lab-scale arms
diff --git a/unilabos/layout_optimizer/models.py b/unilabos/layout_optimizer/models.py
index 7f3e8f99..394c4ef7 100644
--- a/unilabos/layout_optimizer/models.py
+++ b/unilabos/layout_optimizer/models.py
@@ -84,6 +84,8 @@ class Constraint:
     params: dict = field(default_factory=dict)
     # 仅 soft 约束使用
     weight: float = 1.0
+    # 优先级等级，影响有效权重的乘数
+    priority: str | None = None  # "critical" | "high" | "normal" | "low"
 
 
 @dataclass
diff --git a/unilabos/layout_optimizer/optimizer.py b/unilabos/layout_optimizer/optimizer.py
index c86ee007..84ea0aa7 100644
--- a/unilabos/layout_optimizer/optimizer.py
+++ b/unilabos/layout_optimizer/optimizer.py
@@ -1,7 +1,7 @@
 """差分进化布局优化器。
 
 编码：N 个设备 → 3N 维向量 [x0, y0, θ0, x1, y1, θ1, ...]
-使用 scipy.optimize.differential_evolution 进行全局优化。
+使用自定义差分进化循环（per-device crossover + θ wrapping）进行全局优化。
 初始布局（Pencil/回退）注入为种群种子个体加速收敛。
 """
 
@@ -9,19 +9,187 @@ from __future__ import annotations
 
 import logging
 import math
-from typing import Any
+from typing import Any, Callable
 
 import numpy as np
-from scipy.optimize import differential_evolution
 
 from .constraints import evaluate_constraints, evaluate_default_hard_constraints
 from .mock_checkers import MockCollisionChecker, MockReachabilityChecker
 from .models import Constraint, Device, Lab, Placement
 from .pencil_integration import generate_initial_layout
+from .seeders import resolve_seeder_params, seed_layout
 
 logger = logging.getLogger(__name__)
 
 
+def _run_de(
+    cost_fn: Callable[[np.ndarray], float],
+    bounds: np.ndarray,
+    init_pop: np.ndarray,
+    maxiter: int,
+    tol: float,
+    atol: float,
+    mutation: tuple[float, float],
+    recombination: float,
+    seed: int | None,
+    n_devices: int,
+    strategy: str = "currenttobest1bin",
+) -> tuple[np.ndarray, float, int]:
+    """自定义差分进化循环。
+
+    特性：
+    - 支持 currenttobest1bin / best1bin 两种策略
+    - Per-device crossover：以设备 (x, y, θ) 三元组为原子单元进行交叉
+    - θ wrapping：交叉后对角度取模 [0, 2π)
+    - Early stopping：最近 20 代改善 < 0.1% 时提前终止
+    - scipy 风格收敛判断：std(costs) <= atol + tol * |best_cost|
+
+    Args:
+        cost_fn: 目标函数 f(x) → float
+        bounds: 边界数组 shape=(ndim, 2)，每行 [low, high]
+        init_pop: 初始种群 shape=(pop_size, ndim)
+        maxiter: 最大迭代代数
+        tol: 相对收敛容差
+        atol: 绝对收敛容差
+        mutation: 变异因子范围 (F_min, F_max)
+        recombination: 交叉概率 CR
+        seed: 随机种子
+        n_devices: 设备数量（用于 per-device crossover）
+        strategy: 变异策略，"currenttobest1bin" 或 "best1bin"
+
+    Returns:
+        (best_vector, best_cost, n_generations)
+    """
+    rng = np.random.default_rng(seed)
+    pop_size, ndim = init_pop.shape
+    lower = bounds[:, 0]
+    upper = bounds[:, 1]
+    f_min, f_max = mutation
+
+    # 评估初始种群适应度
+    costs = np.array([cost_fn(ind) for ind in init_pop])
+    best_idx = int(np.argmin(costs))
+    best_cost = costs[best_idx]
+    best_vector = init_pop[best_idx].copy()
+
+    # Early stopping 跟踪
+    patience = 20
+    best_cost_history: list[float] = [best_cost]
+
+    for gen in range(1, maxiter + 1):
+        for i in range(pop_size):
+            # 选择变异因子 F（每个个体独立采样）
+            f_val = rng.uniform(f_min, f_max)
+
+            # 选择两个不同于 i 和 best_idx 的个体索引
+            candidates = list(range(pop_size))
+            candidates.remove(i)
+            chosen = rng.choice(candidates, size=2, replace=False)
+            r1, r2 = int(chosen[0]), int(chosen[1])
+
+            # 变异向量
+            if strategy == "best1bin":
+                # Turbo 模式：mutant = best + F*(r1 - r2)
+                mutant = best_vector + f_val * (init_pop[r1] - init_pop[r2])
+            else:
+                # 默认 currenttobest1bin：mutant = target + F*(best - target) + F*(r1 - r2)
+                mutant = (
+                    init_pop[i]
+                    + f_val * (best_vector - init_pop[i])
+                    + f_val * (init_pop[r1] - init_pop[r2])
+                )
+
+            # Per-device crossover：以 (x, y, θ) 三元组为原子单元
+            trial = init_pop[i].copy()
+            j_rand = rng.integers(0, n_devices)  # 保证至少一个设备来自 mutant
+            for d in range(n_devices):
+                if rng.random() < recombination or d == j_rand:
+                    trial[3 * d: 3 * d + 3] = mutant[3 * d: 3 * d + 3]
+
+            # θ wrapping：角度取模 [0, 2π)
+            for d in range(n_devices):
+                trial[3 * d + 2] %= 2 * math.pi
+
+            # 钳位到边界内
+            trial = np.clip(trial, lower, upper)
+
+            # 贪心选择：trial 不比当前差则替换
+            trial_cost = cost_fn(trial)
+            if trial_cost <= costs[i]:
+                init_pop[i] = trial
+                costs[i] = trial_cost
+                if trial_cost < best_cost:
+                    best_cost = trial_cost
+                    best_vector = trial.copy()
+
+        # 更新 best_idx（种群可能整体更新）
+        best_idx = int(np.argmin(costs))
+
+        # Early stopping：最近 patience 代改善 < 0.1%
+        best_cost_history.append(best_cost)
+        if len(best_cost_history) >= patience:
+            old_cost = best_cost_history[-patience]
+            if old_cost > 0:
+                improvement = (old_cost - best_cost) / old_cost
+            else:
+                improvement = 0.0
+            if improvement < 0.001:
+                logger.info(
+                    "Early stop: cost 在 %d 代内稳定在 %.4f（改善 < 0.1%%）",
+                    patience, best_cost,
+                )
+                return best_vector, best_cost, gen
+
+        # scipy 风格收敛判断
+        if np.std(costs) <= atol + tol * abs(best_cost):
+            logger.info(
+                "收敛终止：std(costs)=%.6f <= atol+tol*|best|=%.6f，第 %d 代",
+                np.std(costs), atol + tol * abs(best_cost), gen,
+            )
+            return best_vector, best_cost, gen
+
+    return best_vector, best_cost, maxiter
+
+
+def _generate_seeds(
+    devices: list[Device],
+    lab: Lab,
+    rng: np.random.Generator,
+    workflow_edges: list[list[str]] | None = None,
+    n_variants: int = 3,
+    sigma_pos_frac: float = 0.05,
+    sigma_theta: float = math.pi / 6,
+) -> list[np.ndarray]:
+    """从多个 seeder preset 生成多样性种子个体 + 变异版本。"""
+    seeds: list[np.ndarray] = []
+    presets = ["compact_outward", "spread_inward"]
+    if workflow_edges:
+        presets.append("workflow_cluster")
+
+    for preset_name in presets:
+        try:
+            params = resolve_seeder_params(preset_name)
+        except ValueError:
+            continue
+        if params is None:
+            continue
+        base_placements = seed_layout(devices, lab, params, workflow_edges)
+        base_vec = _placements_to_vector(base_placements, devices)
+        seeds.append(base_vec)
+
+        # 变异版本：对 (x,y) 加高斯噪声 σ=5% lab 尺寸，θ 加 σ=π/6
+        for _ in range(n_variants):
+            variant = base_vec.copy()
+            for d in range(len(devices)):
+                variant[3 * d] += rng.normal(0, sigma_pos_frac * lab.width)
+                variant[3 * d + 1] += rng.normal(0, sigma_pos_frac * lab.depth)
+                variant[3 * d + 2] += rng.normal(0, sigma_theta)
+                variant[3 * d + 2] %= 2 * math.pi
+            seeds.append(variant)
+
+    return seeds
+
+
 def optimize(
     devices: list[Device],
     lab: Lab,
@@ -33,6 +201,8 @@ def optimize(
     popsize: int = 15,
     tol: float = 1e-6,
     seed: int | None = None,
+    strategy: str = "currenttobest1bin",
+    workflow_edges: list[list[str]] | None = None,
 ) -> list[Placement]:
     """运行差分进化优化，返回最优布局。
 
@@ -47,6 +217,7 @@ def optimize(
         popsize: 种群大小倍数
         tol: 收敛容差
         seed: 随机种子（用于可复现性）
+        strategy: DE 变异策略（"currenttobest1bin" 或 "best1bin"）
 
     Returns:
         最优布局 Placement 列表
@@ -105,57 +276,49 @@ def optimize(
 
         return hard_cost
 
-    # 构建初始种群：种子个体 + 随机个体
+    # 构建初始种群：种子个体 + 多样性种子 + 随机个体
     rng = np.random.default_rng(seed)
     pop_count = popsize * 3 * n  # scipy 默认 popsize * dim
     init_pop = rng.uniform(
         bounds_array[:, 0], bounds_array[:, 1], size=(pop_count, 3 * n)
     )
-    init_pop[0] = seed_vector  # 注入种子
+    init_pop[0] = seed_vector  # 注入原始种子
+
+    # 多样性种子注入（多 preset + 变异版本）
+    extra_seeds = _generate_seeds(devices, lab, rng, workflow_edges)
+    for i, s in enumerate(extra_seeds):
+        idx = i + 1  # 原始种子占 [0]
+        if idx < pop_count:
+            init_pop[idx] = np.clip(s, bounds_array[:, 0], bounds_array[:, 1])
 
     logger.info(
-        "Starting DE optimization: %d devices, %d-dim, popsize=%d, maxiter=%d",
-        n, 3 * n, pop_count, maxiter,
+        "Starting DE optimization: %d devices, %d-dim, popsize=%d, maxiter=%d, strategy=%s",
+        n, 3 * n, pop_count, maxiter, strategy,
     )
 
-    # Early stopping: stop when cost hasn't improved by >0.1% for 20 generations
-    _best_costs: list[float] = []
-    _patience = 20
-
-    def _early_stop_callback(xk, convergence=0):
-        cost = cost_function(xk)
-        _best_costs.append(cost)
-        if len(_best_costs) >= _patience:
-            recent = _best_costs[-_patience:]
-            if recent[0] > 0:
-                improvement = (recent[0] - recent[-1]) / recent[0]
-            else:
-                improvement = 0.0
-            if improvement < 0.001:  # < 0.1% improvement over last 20 gens
-                logger.info("Early stop: cost stable at %.4f for %d generations", cost, _patience)
-                return True
-        return False
-
-    result = differential_evolution(
-        cost_function,
-        bounds=list(bounds),
-        init=init_pop,
+    best_vector, best_cost, n_generations = _run_de(
+        cost_fn=cost_function,
+        bounds=bounds_array,
+        init_pop=init_pop,
         maxiter=maxiter,
         tol=tol,
         atol=1e-3,
         mutation=(0.5, 1.0),
         recombination=0.7,
         seed=seed,
-        disp=False,
-        callback=_early_stop_callback,
+        n_devices=n,
+        strategy=strategy,
     )
 
+    # 评估次数估算：每代 pop_count 次（初始 + 每代 trial）
+    n_evaluations = pop_count + n_generations * pop_count
+
     logger.info(
         "DE optimization complete: success=%s, cost=%.4f, iterations=%d, evaluations=%d",
-        result.success, result.fun, result.nit, result.nfev,
+        best_cost < 1e17, best_cost, n_generations, n_evaluations,
     )
 
-    return _vector_to_placements(result.x, devices)
+    return _vector_to_placements(best_vector, devices)
 
 
 def snap_theta(placements: list[Placement], threshold_deg: float = 15.0) -> list[Placement]:
@@ -179,6 +342,38 @@ def snap_theta(placements: list[Placement], threshold_deg: float = 15.0) -> list
     return result
 
 
+def snap_theta_safe(
+    placements: list[Placement],
+    devices: list[Device],
+    lab: Lab,
+    collision_checker: Any,
+    threshold_deg: float = 15.0,
+) -> list[Placement]:
+    """Snap theta 到基数方向，但碰撞时回退到原始角度。
+
+    逐设备检查：snap 后如果产生碰撞或越界，则该设备保留原始 theta。
+    """
+    snapped = snap_theta(placements, threshold_deg)
+
+    result = list(snapped)
+    for idx, (orig, snap) in enumerate(zip(placements, snapped)):
+        if abs(orig.theta - snap.theta) < 1e-9:
+            continue  # 未 snap，跳过
+        # 检查 snap 版本是否导致新碰撞
+        test_placements = result.copy()
+        test_placements[idx] = snap
+        cost = evaluate_default_hard_constraints(
+            devices, test_placements, lab, collision_checker, graduated=False,
+        )
+        if math.isinf(cost):
+            result[idx] = orig  # 回退到未 snap 的角度
+            logger.info(
+                "snap_theta_safe: 设备 %s snap θ=%.2f→%.2f 导致碰撞，已回退",
+                snap.device_id, orig.theta, snap.theta,
+            )
+    return result
+
+
 def _placements_to_vector(
     placements: list[Placement], devices: list[Device]
 ) -> np.ndarray:
@@ -208,7 +403,7 @@ def _vector_to_placements(
                 device_id=dev.id,
                 x=float(x[3 * i]),
                 y=float(x[3 * i + 1]),
-                theta=float(x[3 * i + 2]),
+                theta=float(x[3 * i + 2] % (2 * math.pi)),
             )
         )
     return placements
diff --git a/unilabos/layout_optimizer/server.py b/unilabos/layout_optimizer/server.py
index caf9cb61..f24f68a6 100644
--- a/unilabos/layout_optimizer/server.py
+++ b/unilabos/layout_optimizer/server.py
@@ -23,6 +23,7 @@ from fastapi.responses import FileResponse, RedirectResponse
 from fastapi.staticfiles import StaticFiles
 from pydantic import BaseModel
 
+from .constraints import DEFAULT_WEIGHT_ANGLE
 from .device_catalog import (
     create_devices_from_list,
     load_devices_from_assets,
@@ -395,9 +396,9 @@ async def run_optimize(request: OptimizeRequest):
     """接收设备列表+约束，返回最优布局方案。"""
     from fastapi import HTTPException
 
-    from .constraints import evaluate_default_hard_constraints
+    from .constraints import evaluate_default_hard_constraints, evaluate_constraints
     from .mock_checkers import MockCollisionChecker
-    from .optimizer import optimize, snap_theta
+    from .optimizer import optimize, snap_theta, snap_theta_safe
     from .seeders import resolve_seeder_params, seed_layout
 
     logger.info(
@@ -456,10 +457,10 @@ async def run_optimize(request: OptimizeRequest):
                 type="soft",
                 rule_name="prefer_orientation_mode",
                 params={"mode": orientation_mode},
-                weight=request.seeder_overrides.get("orientation_weight", 5.0),
+                weight=request.seeder_overrides.get("orientation_weight", DEFAULT_WEIGHT_ANGLE),
             ))
         # prefer_aligned: penalize non-cardinal angles
-        align_weight = request.seeder_overrides.get("align_weight", 2.0)
+        align_weight = request.seeder_overrides.get("align_weight", DEFAULT_WEIGHT_ANGLE)
         if align_weight > 0:
             constraints.append(Constraint(
                 type="soft",
@@ -479,18 +480,27 @@ async def run_optimize(request: OptimizeRequest):
             seed_placements=seed_placements,
             maxiter=request.maxiter,
             seed=request.seed,
+            workflow_edges=request.workflow_edges or None,
         )
         de_ran = True
     else:
         result_placements = seed_placements
 
-    # 5. θ snap post-processing
-    result_placements = snap_theta(result_placements)
+    # 5. θ snap post-processing（碰撞安全：snap 后验证，失败则回退）
+    result_placements = snap_theta_safe(result_placements, devices, lab, checker)
 
     # 6. Evaluate final cost (binary mode for pass/fail reporting)
     final_cost = evaluate_default_hard_constraints(
         devices, result_placements, lab, checker, graduated=False,
     )
+    # 也检查用户硬约束（binary 模式）
+    if constraints and not math.isinf(final_cost):
+        user_hard_cost = evaluate_constraints(
+            devices, result_placements, lab, constraints, checker,
+            graduated=False,
+        )
+        if math.isinf(user_hard_cost):
+            final_cost = math.inf
 
     return OptimizeResponse(
         placements=[
diff --git a/unilabos/layout_optimizer/tests/test_broad_phase.py b/unilabos/layout_optimizer/tests/test_broad_phase.py
new file mode 100644
index 00000000..444a4e40
--- /dev/null
+++ b/unilabos/layout_optimizer/tests/test_broad_phase.py
@@ -0,0 +1,241 @@
+"""Tests for broad_phase.py — 2 轴 sweep-and-prune 宽相碰撞检测。"""
+
+from __future__ import annotations
+
+import math
+import random
+
+import pytest
+
+from ..broad_phase import broad_phase_device_pairs, sweep_and_prune_pairs
+from ..models import Device, Placement
+
+
+# ---------------------------------------------------------------------------
+# 测试用辅助函数
+# ---------------------------------------------------------------------------
+
+def _make_device(device_id: str, w: float = 0.6, d: float = 0.4) -> Device:
+    """创建简单测试设备。"""
+    return Device(id=device_id, name=device_id, bbox=(w, d))
+
+
+def _make_placement(
+    device_id: str, x: float, y: float, theta: float = 0.0
+) -> Placement:
+    """创建简单测试放置。"""
+    return Placement(device_id=device_id, x=x, y=y, theta=theta)
+
+
+# ---------------------------------------------------------------------------
+# 测试类
+# ---------------------------------------------------------------------------
+
+
+class TestNoOverlap:
+    """两台设备距离足够远，宽相不返回候选对。"""
+
+    def test_no_overlap_returns_empty(self):
+        """水平方向间距远大于 AABB 尺寸 → 0 候选对。"""
+        devices = [_make_device("A", 1.0, 1.0), _make_device("B", 1.0, 1.0)]
+        placements = [
+            _make_placement("A", 0.0, 0.0),
+            _make_placement("B", 10.0, 0.0),
+        ]
+        pairs = sweep_and_prune_pairs(devices, placements)
+        assert pairs == []
+
+
+class TestOverlapping:
+    """两台设备 AABB 明显重叠。"""
+
+    def test_overlapping_devices_returned(self):
+        """两台 1×1 设备中心距 0.5m → 1 候选对。"""
+        devices = [_make_device("A", 1.0, 1.0), _make_device("B", 1.0, 1.0)]
+        placements = [
+            _make_placement("A", 0.0, 0.0),
+            _make_placement("B", 0.5, 0.0),
+        ]
+        pairs = sweep_and_prune_pairs(devices, placements)
+        assert len(pairs) == 1
+        assert pairs[0] == (0, 1)
+
+
+class TestXOverlapYNoOverlap:
+    """x 轴投影交叠但 y 轴不交叠，应被 y 轴检查剪枝。"""
+
+    def test_x_overlap_y_no_overlap(self):
+        """水平接近但垂直方向偏移足够大 → 0 候选对。"""
+        devices = [_make_device("A", 2.0, 1.0), _make_device("B", 2.0, 1.0)]
+        placements = [
+            _make_placement("A", 0.0, 0.0),
+            _make_placement("B", 0.5, 5.0),  # x 轴交叠但 y 轴相距很远
+        ]
+        pairs = sweep_and_prune_pairs(devices, placements)
+        assert pairs == []
+
+
+class TestTouchingDevices:
+    """AABB 恰好接触（边缘距离 = 0）应作为候选对返回。"""
+
+    def test_touching_devices_included(self):
+        """两个 1×1 设备中心距恰好为 1.0（半宽 0.5 + 0.5 = 1.0），
+        AABB 边界接触 → 应包含在候选对中（<= 判定）。"""
+        devices = [_make_device("A", 1.0, 1.0), _make_device("B", 1.0, 1.0)]
+        placements = [
+            _make_placement("A", 0.0, 0.0),
+            _make_placement("B", 1.0, 0.0),  # xmax_A = 0.5, xmin_B = 0.5 → 接触
+        ]
+        pairs = sweep_and_prune_pairs(devices, placements)
+        # 接触算作潜在碰撞，安全起见需保留
+        assert len(pairs) == 1
+        assert pairs[0] == (0, 1)
+
+
+class TestMultipleDevices:
+    """4 台设备验证精确的候选对列表。"""
+
+    def test_multiple_devices_correct_pairs(self):
+        """排列 4 台设备，只有特定配对 AABB 交叠。
+
+        布局（1×1 设备）：
+          A(0,0)  B(0.8,0)  — A-B 交叠（中心距 0.8 < 1.0）
+          C(0,5)            — 远离 A、B
+          D(0.9,5)          — C-D 交叠（中心距 0.9 < 1.0）
+
+        期望候选对: (A,B) 和 (C,D)。
+        """
+        devices = [
+            _make_device("A", 1.0, 1.0),
+            _make_device("B", 1.0, 1.0),
+            _make_device("C", 1.0, 1.0),
+            _make_device("D", 1.0, 1.0),
+        ]
+        placements = [
+            _make_placement("A", 0.0, 0.0),
+            _make_placement("B", 0.8, 0.0),
+            _make_placement("C", 0.0, 5.0),
+            _make_placement("D", 0.9, 5.0),
+        ]
+        pairs = sweep_and_prune_pairs(devices, placements)
+        pair_set = set(pairs)
+        assert (0, 1) in pair_set  # A-B
+        assert (2, 3) in pair_set  # C-D
+        assert len(pair_set) == 2
+
+
+class TestRotatedDeviceAabb:
+    """旋转设备导致 AABB 变大，命中候选对。"""
+
+    def test_rotated_device_aabb(self):
+        """一台窄长设备 (2.0×0.2)：
+        - 未旋转时 AABB 半宽 = 1.0，两设备中心距 2.5 → 不交叠
+        - 旋转 90° 后 AABB 半宽 = 0.1，半深 = 1.0 → 仍不交叠
+        - 旋转 45° 后 AABB 半宽 ≈ (2*cos45 + 0.2*sin45)/2 ≈ 0.778
+          但另一台放在 x=1.6，半宽 = 1.0
+          所以 xmax_A = 0 + 0.778 = 0.778 < 1.6 - 1.0 = 0.6 → 不够
+
+        更好的方案：用中心距 1.5，未旋转时不交叠，旋转后交叠。
+        未旋转: half_w_A = 0.3 (bbox 0.6x2.0), half_w_B = 0.3
+          A: xmax = 0 + 0.3 = 0.3, B: xmin = 1.5 - 0.3 = 1.2 → 不交叠
+        旋转 90°: A 的 bbox (0.6, 2.0) → half_w = (0.6*0 + 2.0*1)/2 = 1.0
+          A: xmax = 0 + 1.0 = 1.0, B: xmin = 1.5 - 0.3 = 1.2 → 仍不交叠
+
+        用 bbox (0.4, 2.0)，间距 1.2：
+        未旋转: half_w = 0.2, xmax_A = 0.2, xmin_B = 1.2 - 0.2 = 1.0 → 不交叠
+        旋转 45°: half_w = (0.4*cos45 + 2.0*sin45)/2 = (0.283+1.414)/2 = 0.849
+          xmax_A = 0.849, xmin_B = 1.2 - 0.2 = 1.0 → 不交叠
+
+        间距 0.8：
+        未旋转: xmax_A = 0.2, xmin_B = 0.8 - 0.2 = 0.6 → 不交叠 ✓
+        旋转 45°: xmax_A = 0.849, xmin_B = 0.6 → 交叠 ✓ (0.849 > 0.6)
+        y 轴: half_d_A_rot = (0.4*sin45 + 2.0*cos45)/2 = 0.849, half_d_B = 1.0
+          ymax_A = 0.849, ymin_B = -1.0 → 交叠 ✓
+        """
+        dev_narrow = _make_device("narrow", 0.4, 2.0)
+        dev_normal = _make_device("normal", 0.4, 2.0)
+        # 未旋转：不交叠
+        placements_no_rot = [
+            _make_placement("narrow", 0.0, 0.0, theta=0.0),
+            _make_placement("normal", 0.8, 0.0, theta=0.0),
+        ]
+        assert sweep_and_prune_pairs([dev_narrow, dev_normal], placements_no_rot) == []
+
+        # narrow 旋转 45° → AABB 变大 → 交叠
+        placements_rot = [
+            _make_placement("narrow", 0.0, 0.0, theta=math.pi / 4),
+            _make_placement("normal", 0.8, 0.0, theta=0.0),
+        ]
+        pairs = sweep_and_prune_pairs([dev_narrow, dev_normal], placements_rot)
+        assert len(pairs) == 1
+
+
+class TestOriginalIndices:
+    """验证返回的索引对应 placements 原始顺序而非排序后顺序。"""
+
+    def test_sorted_output_preserves_original_indices(self):
+        """故意让 placements 按 x 坐标逆序排列，
+        验证返回的索引仍是原始顺序。"""
+        devices = [
+            _make_device("A", 1.0, 1.0),
+            _make_device("B", 1.0, 1.0),
+            _make_device("C", 1.0, 1.0),
+        ]
+        # 逆序排列：C 在最左，A 在最右
+        placements = [
+            _make_placement("A", 5.0, 0.0),  # idx 0, 最右
+            _make_placement("B", 4.5, 0.0),  # idx 1, 中间（与 A 交叠）
+            _make_placement("C", 0.0, 0.0),  # idx 2, 最左（独立）
+        ]
+        pairs = sweep_and_prune_pairs(devices, placements)
+        # A(idx=0) 和 B(idx=1) AABB 交叠，索引应为 (0, 1)
+        assert len(pairs) == 1
+        assert pairs[0] == (0, 1)
+
+        # 同时验证 broad_phase_device_pairs 返回正确 device_id
+        id_pairs = broad_phase_device_pairs(devices, placements)
+        assert id_pairs == [("A", "B")]
+
+
+class TestPairCountReduction:
+    """大规模随机测试：宽相候选对数应远小于 N*(N-1)/2。"""
+
+    def test_pair_count_reduction(self):
+        """N=15 台设备随机放置在 10×10 实验室 → 候选对数显著少于全量。"""
+        random.seed(42)
+        n = 15
+        devices = [_make_device(f"D{i}", 0.5, 0.5) for i in range(n)]
+        placements = [
+            _make_placement(f"D{i}", random.uniform(0, 10), random.uniform(0, 10))
+            for i in range(n)
+        ]
+        pairs = sweep_and_prune_pairs(devices, placements)
+        full_pairs = n * (n - 1) // 2  # = 105
+        # 在 10×10 区域放 15 台 0.5×0.5 设备，交叠率应很低
+        assert len(pairs) < full_pairs
+        # 额外断言：候选对数不超过全量的一半（保守判定）
+        assert len(pairs) < full_pairs * 0.5
+
+
+class TestEdgeCases:
+    """边界情况。"""
+
+    def test_empty_input(self):
+        """空列表 → 空结果。"""
+        assert sweep_and_prune_pairs([], []) == []
+
+    def test_single_device(self):
+        """单台设备 → 无候选对。"""
+        devices = [_make_device("A")]
+        placements = [_make_placement("A", 0.0, 0.0)]
+        assert sweep_and_prune_pairs(devices, placements) == []
+
+    def test_identical_positions(self):
+        """两台设备完全重叠 → 1 候选对。"""
+        devices = [_make_device("A", 1.0, 1.0), _make_device("B", 1.0, 1.0)]
+        placements = [
+            _make_placement("A", 0.0, 0.0),
+            _make_placement("B", 0.0, 0.0),
+        ]
+        pairs = sweep_and_prune_pairs(devices, placements)
+        assert len(pairs) == 1
diff --git a/unilabos/layout_optimizer/tests/test_bugfixes_v2.py b/unilabos/layout_optimizer/tests/test_bugfixes_v2.py
index 021ebd6f..8e63fa31 100644
--- a/unilabos/layout_optimizer/tests/test_bugfixes_v2.py
+++ b/unilabos/layout_optimizer/tests/test_bugfixes_v2.py
@@ -90,9 +90,16 @@ class TestDuplicateDeviceIDs:
         lab = Lab(width=5, depth=5)
         constraints = [Constraint(type="hard", rule_name="min_spacing",
                                   params={"min_gap": 0.05})]
+        # graduated=True (default): 返回有限惩罚
         cost = evaluate_constraints(devices, stacked, lab, constraints,
                                     MockCollisionChecker())
-        assert math.isinf(cost)
+        assert cost > 0
+        assert not math.isinf(cost)
+        # graduated=False: binary inf
+        cost_binary = evaluate_constraints(devices, stacked, lab, constraints,
+                                           MockCollisionChecker(),
+                                           graduated=False)
+        assert math.isinf(cost_binary)
 
     def test_create_devices_uses_uuid(self):
         """create_devices_from_list should use uuid as Device.id."""
diff --git a/unilabos/layout_optimizer/tests/test_constraints.py b/unilabos/layout_optimizer/tests/test_constraints.py
index 66ba04fd..ac8b3a68 100644
--- a/unilabos/layout_optimizer/tests/test_constraints.py
+++ b/unilabos/layout_optimizer/tests/test_constraints.py
@@ -123,7 +123,7 @@ class TestUserConstraints:
         assert cost == 0.0
 
     def test_distance_less_than_violated_hard(self):
-        """硬距离约束违反返回 inf。"""
+        """硬距离约束违反：graduated模式返回有限惩罚，binary模式返回inf。"""
         devices = _make_devices()
         placements = [
             Placement("a", 1.0, 1.0, 0.0),
@@ -134,10 +134,18 @@ class TestUserConstraints:
                        params={"device_a": "a", "device_b": "b", "distance": 1.0})
         ]
         checker = MockCollisionChecker()
+        # graduated=True (default): 有限惩罚
         cost = evaluate_constraints(
             devices, placements, _make_lab(), constraints, checker
         )
-        assert math.isinf(cost)
+        assert cost > 0
+        assert not math.isinf(cost)
+        # graduated=False: binary inf
+        cost_binary = evaluate_constraints(
+            devices, placements, _make_lab(), constraints, checker,
+            graduated=False,
+        )
+        assert math.isinf(cost_binary)
 
     def test_minimize_distance_cost(self):
         """minimize_distance 约束应返回正比于距离的 cost。"""
@@ -184,7 +192,7 @@ class TestUserConstraints:
         assert not math.isinf(cost)  # reachable → no hard failure
 
     def test_reachability_constraint_violated(self):
-        """可达性约束：目标超出臂展应返回 inf。"""
+        """可达性约束：目标超出臂展 — graduated返回有限惩罚，binary返回inf。"""
         devices = [
             Device(id="arm", name="Arm", bbox=(0.2, 0.2), device_type="articulation"),
             Device(id="target", name="Target", bbox=(0.5, 0.5)),
@@ -199,10 +207,18 @@ class TestUserConstraints:
         ]
         checker = MockCollisionChecker()
         reachability = MockReachabilityChecker(arm_reach={"arm": 1.0})
+        # graduated=True (default): 有限惩罚
         cost = evaluate_constraints(
             devices, placements, _make_lab(), constraints, checker, reachability
         )
-        assert math.isinf(cost)
+        assert cost > 0
+        assert not math.isinf(cost)
+        # graduated=False: binary inf
+        cost_binary = evaluate_constraints(
+            devices, placements, _make_lab(), constraints, checker, reachability,
+            graduated=False,
+        )
+        assert math.isinf(cost_binary)
 
 
 def test_distance_less_than_uses_edge_to_edge():
@@ -272,3 +288,136 @@ def test_prefer_aligned_sums_over_devices():
     cost = evaluate_constraints(devices, placements, lab, [constraint], checker)
     # 2 devices × 1.0 × weight 2.0 = 4.0
     assert cost == pytest.approx(4.0)
+
+
+class TestGraduatedHardConstraints:
+    """graduated 模式下硬约束返回比例惩罚而非 inf。"""
+
+    def test_hard_reachability_graduated_finite(self):
+        """graduated=True: 硬可达性返回有限惩罚。"""
+        devices = [
+            Device(id="arm", name="Arm", bbox=(0.2, 0.2), device_type="articulation"),
+            Device(id="t", name="Target", bbox=(0.5, 0.5)),
+        ]
+        placements = [
+            Placement("arm", 1.0, 1.0, 0.0),
+            Placement("t", 4.0, 3.0, 0.0),
+        ]
+        constraints = [
+            Constraint(type="hard", rule_name="reachability",
+                       params={"arm_id": "arm", "target_device_id": "t"}, weight=1.0)
+        ]
+        checker = MockCollisionChecker()
+        reach = MockReachabilityChecker(arm_reach={"arm": 1.0})
+        cost = evaluate_constraints(
+            devices, placements, _make_lab(), constraints, checker, reach,
+            graduated=True,
+        )
+        assert cost > 0
+        assert not math.isinf(cost)
+
+    def test_hard_reachability_binary_inf(self):
+        """graduated=False: 硬可达性返回 inf。"""
+        devices = [
+            Device(id="arm", name="Arm", bbox=(0.2, 0.2), device_type="articulation"),
+            Device(id="t", name="Target", bbox=(0.5, 0.5)),
+        ]
+        placements = [
+            Placement("arm", 1.0, 1.0, 0.0),
+            Placement("t", 4.0, 3.0, 0.0),
+        ]
+        constraints = [
+            Constraint(type="hard", rule_name="reachability",
+                       params={"arm_id": "arm", "target_device_id": "t"}, weight=1.0)
+        ]
+        checker = MockCollisionChecker()
+        reach = MockReachabilityChecker(arm_reach={"arm": 1.0})
+        cost = evaluate_constraints(
+            devices, placements, _make_lab(), constraints, checker, reach,
+            graduated=False,
+        )
+        assert math.isinf(cost)
+
+    def test_hard_min_spacing_graduated_sums_all_pairs(self):
+        """graduated模式：min_spacing 对所有违规对求和（不只第一对）。"""
+        devices = [
+            Device(id="a", name="A", bbox=(0.5, 0.5)),
+            Device(id="b", name="B", bbox=(0.5, 0.5)),
+            Device(id="c", name="C", bbox=(0.5, 0.5)),
+        ]
+        # 三个设备间距都小于 min_gap=1.0
+        placements = [
+            Placement("a", 1.0, 2.0, 0.0),
+            Placement("b", 1.3, 2.0, 0.0),  # OBB 边缘距 a 约 0.3
+            Placement("c", 1.6, 2.0, 0.0),  # OBB 边缘距 b 约 0.3, 距 a 约 0.6
+        ]
+        constraints = [
+            Constraint(type="hard", rule_name="min_spacing",
+                       params={"min_gap": 1.0}, weight=1.0)
+        ]
+        checker = MockCollisionChecker()
+        cost = evaluate_constraints(
+            devices, placements, _make_lab(), constraints, checker,
+            graduated=True,
+        )
+        # 应大于 0 且有限（累加多对违规）
+        assert cost > 0
+        assert not math.isinf(cost)
+
+    def test_hard_min_spacing_binary_inf(self):
+        """graduated=False: min_spacing 违规返回 inf。"""
+        devices = _make_devices()
+        placements = [
+            Placement("a", 1.0, 2.0, 0.0),
+            Placement("b", 1.3, 2.0, 0.0),
+        ]
+        constraints = [
+            Constraint(type="hard", rule_name="min_spacing",
+                       params={"min_gap": 1.0}, weight=1.0)
+        ]
+        checker = MockCollisionChecker()
+        cost = evaluate_constraints(
+            devices, placements, _make_lab(), constraints, checker,
+            graduated=False,
+        )
+        assert math.isinf(cost)
+
+    def test_hard_distance_less_than_graduated(self):
+        """graduated模式：distance_less_than 硬约束返回比例惩罚。"""
+        devices = _make_devices()
+        placements = [
+            Placement("a", 1.0, 2.0, 0.0),
+            Placement("b", 4.0, 2.0, 0.0),
+        ]
+        constraints = [
+            Constraint(type="hard", rule_name="distance_less_than",
+                       params={"device_a": "a", "device_b": "b", "distance": 0.5},
+                       weight=2.0)
+        ]
+        checker = MockCollisionChecker()
+        cost = evaluate_constraints(
+            devices, placements, _make_lab(), constraints, checker,
+            graduated=True,
+        )
+        # HARD_MULTIPLIER(5) × weight(2) × overshoot > 0
+        assert cost > 0
+        assert not math.isinf(cost)
+
+    def test_graduated_default_is_true(self):
+        """不传 graduated 参数时默认使用 graduated 模式。"""
+        devices = _make_devices()
+        placements = [
+            Placement("a", 1.0, 2.0, 0.0),
+            Placement("b", 4.0, 2.0, 0.0),
+        ]
+        constraints = [
+            Constraint(type="hard", rule_name="distance_less_than",
+                       params={"device_a": "a", "device_b": "b", "distance": 0.5},
+                       weight=1.0)
+        ]
+        checker = MockCollisionChecker()
+        # 不指定 graduated — 默认应为 True → 有限惩罚
+        cost = evaluate_constraints(
+            devices, placements, _make_lab(), constraints, checker,
+        )
+        assert not math.isinf(cost)
diff --git a/unilabos/layout_optimizer/tests/test_e2e_pcr_pipeline.py b/unilabos/layout_optimizer/tests/test_e2e_pcr_pipeline.py
index 5910d186..53557722 100644
--- a/unilabos/layout_optimizer/tests/test_e2e_pcr_pipeline.py
+++ b/unilabos/layout_optimizer/tests/test_e2e_pcr_pipeline.py
@@ -185,9 +185,9 @@ class TestStage3VerifyPlacements:
     def test_no_hard_constraint_violation(self):
         """Full pipeline with all intents including reachability converges cleanly.
 
-        MockReachabilityChecker uses large fallback reach for unknown arms,
-        so arm_slider reachability constraints are satisfied in mock mode.
-        When real ROS checkers replace mock, this test validates the same pipeline.
+        MockReachabilityChecker now includes arm_slider in the default reach table
+        (1.07m). Binary final evaluation checks all hard constraints including
+        user-defined reachability.
         """
         interpret_data = client.post("/interpret", json={"intents": LLM_INTENTS}).json()
 
@@ -197,7 +197,7 @@ class TestStage3VerifyPlacements:
             "constraints": interpret_data["constraints"],
             "workflow_edges": interpret_data["workflow_edges"],
             "run_de": True,
-            "maxiter": 50,
+            "maxiter": 100,
             "seed": 42,
         })
         data = optimize_resp.json()
diff --git a/unilabos/layout_optimizer/tests/test_optimizer.py b/unilabos/layout_optimizer/tests/test_optimizer.py
index 3d430bfa..a7892092 100644
--- a/unilabos/layout_optimizer/tests/test_optimizer.py
+++ b/unilabos/layout_optimizer/tests/test_optimizer.py
@@ -4,8 +4,9 @@ import math
 
 from ..mock_checkers import MockCollisionChecker
 from ..models import Device, Lab, Placement
+import numpy as np
 import pytest
-from ..optimizer import optimize, snap_theta
+from ..optimizer import _run_de, optimize, snap_theta
 
 
 def test_optimize_three_devices_no_collision():
@@ -217,3 +218,224 @@ def test_full_pipeline_with_de():
     for p in result:
         assert 0 <= p.x <= lab.width
         assert 0 <= p.y <= lab.depth
+
+
+# ──────────────────────────────────────────────────────────────
+# DE V2 新增测试
+# ──────────────────────────────────────────────────────────────
+
+
+def test_theta_wrapping_convergence():
+    """θ 在 0/2π 边界附近应正确收敛，不发散。
+
+    构造一个简单的 cost function，最优解 θ≈0（即 2π 附近也可以）。
+    验证自定义 DE 在 θ wrapping 下能收敛。
+    """
+    n_devices = 2
+
+    def cost_fn(x: np.ndarray) -> float:
+        # 最优：两设备分开放置，θ 接近 0
+        cost = 0.0
+        for d in range(n_devices):
+            theta = x[3 * d + 2]
+            # θ penalty: 偏离 0 的圆周距离
+            cost += (1 - math.cos(theta)) * 10.0
+        # 两设备距离 penalty
+        dx = x[0] - x[3]
+        dy = x[1] - x[4]
+        dist = math.sqrt(dx * dx + dy * dy)
+        if dist < 1.0:
+            cost += (1.0 - dist) * 100.0
+        return cost
+
+    bounds = np.array([
+        [0.5, 4.5], [0.5, 4.5], [0.0, 2 * math.pi],  # 设备 0
+        [0.5, 4.5], [0.5, 4.5], [0.0, 2 * math.pi],  # 设备 1
+    ])
+    rng = np.random.default_rng(42)
+    pop_size = 30
+    init_pop = rng.uniform(bounds[:, 0], bounds[:, 1], size=(pop_size, 6))
+    # 故意注入 θ 接近 2π 的个体，测试 wrapping
+    init_pop[0, 2] = 2 * math.pi - 0.01
+    init_pop[0, 5] = 2 * math.pi - 0.05
+
+    best_vec, best_cost, n_gen = _run_de(
+        cost_fn=cost_fn,
+        bounds=bounds,
+        init_pop=init_pop,
+        maxiter=100,
+        tol=1e-6,
+        atol=1e-3,
+        mutation=(0.5, 1.0),
+        recombination=0.7,
+        seed=42,
+        n_devices=n_devices,
+    )
+
+    # θ 应在 [0, 2π) 范围内
+    for d in range(n_devices):
+        theta = best_vec[3 * d + 2]
+        assert 0 <= theta < 2 * math.pi, f"θ 超出 [0, 2π): {theta}"
+
+    # cost 应显著下降（不发散）
+    assert best_cost < 50.0, f"θ wrapping 未正确收敛，cost={best_cost}"
+
+
+def test_per_device_crossover_atomicity():
+    """验证 per-device crossover 的原子性：同一设备的 (x, y, θ) 来自同一来源。
+
+    构造 2 设备场景，跟踪 crossover 后每个设备三元组是否完整来自 mutant 或 parent。
+    """
+    rng = np.random.default_rng(123)
+    n_devices = 3
+    ndim = 3 * n_devices
+
+    # 构造明显不同的 parent 和 mutant
+    parent = np.zeros(ndim)
+    mutant = np.ones(ndim) * 10.0
+
+    # 模拟多次 per-device crossover，检查原子性
+    violations = 0
+    n_trials = 200
+    for _ in range(n_trials):
+        trial = parent.copy()
+        j_rand = rng.integers(0, n_devices)
+        for d in range(n_devices):
+            if rng.random() < 0.7 or d == j_rand:
+                trial[3 * d: 3 * d + 3] = mutant[3 * d: 3 * d + 3]
+
+        # 检查每个设备的三元组要么全是 0（parent），要么全是 10（mutant）
+        for d in range(n_devices):
+            triple = trial[3 * d: 3 * d + 3]
+            all_parent = np.allclose(triple, 0.0)
+            all_mutant = np.allclose(triple, 10.0)
+            if not (all_parent or all_mutant):
+                violations += 1
+
+    assert violations == 0, f"Per-device crossover 原子性违反 {violations}/{n_trials * n_devices} 次"
+
+
+def test_strategy_currenttobest1bin_converges():
+    """currenttobest1bin 策略在简单 2 设备问题上应能收敛。"""
+    devices = [
+        Device(id="a", name="A", bbox=(0.6, 0.4)),
+        Device(id="b", name="B", bbox=(0.6, 0.4)),
+    ]
+    lab = Lab(width=5.0, depth=5.0)
+
+    placements = optimize(
+        devices, lab, seed=42, maxiter=80, popsize=10,
+        strategy="currenttobest1bin",
+    )
+
+    assert len(placements) == 2
+    checker = MockCollisionChecker()
+    checker_placements = [
+        {"id": p.device_id, "bbox": next(d.bbox for d in devices if d.id == p.device_id),
+         "pos": (p.x, p.y, p.theta)}
+        for p in placements
+    ]
+    collisions = checker.check(checker_placements)
+    assert collisions == [], f"currenttobest1bin 策略产生碰撞: {collisions}"
+
+
+def test_strategy_best1bin_converges():
+    """best1bin 策略在简单 2 设备问题上应能收敛。"""
+    devices = [
+        Device(id="a", name="A", bbox=(0.6, 0.4)),
+        Device(id="b", name="B", bbox=(0.6, 0.4)),
+    ]
+    lab = Lab(width=5.0, depth=5.0)
+
+    placements = optimize(
+        devices, lab, seed=42, maxiter=80, popsize=10,
+        strategy="best1bin",
+    )
+
+    assert len(placements) == 2
+    checker = MockCollisionChecker()
+    checker_placements = [
+        {"id": p.device_id, "bbox": next(d.bbox for d in devices if d.id == p.device_id),
+         "pos": (p.x, p.y, p.theta)}
+        for p in placements
+    ]
+    collisions = checker.check(checker_placements)
+    assert collisions == [], f"best1bin 策略产生碰撞: {collisions}"
+
+
+def test_convergence_quality_two_devices():
+    """2 设备无碰撞放置，cost 应低于阈值，验证优化质量。"""
+    devices = [
+        Device(id="d1", name="D1", bbox=(0.8, 0.6)),
+        Device(id="d2", name="D2", bbox=(0.6, 0.5)),
+    ]
+    lab = Lab(width=5.0, depth=5.0)
+
+    placements = optimize(devices, lab, seed=42, maxiter=100, popsize=10)
+
+    # 验证结果质量：cost 应很低（无碰撞、在边界内）
+    from ..constraints import evaluate_default_hard_constraints
+    checker = MockCollisionChecker()
+    cost = evaluate_default_hard_constraints(devices, placements, lab, checker)
+    assert cost < 1.0, f"优化质量不佳，cost={cost}"
+
+
+def test_run_de_early_stopping():
+    """验证 _run_de 的 early stopping 机制：简单 cost function 应提前终止。"""
+
+    def trivial_cost(x: np.ndarray) -> float:
+        # 简单二次函数，最优解在原点附近
+        return float(np.sum(x ** 2))
+
+    ndim = 6
+    n_devices = 2
+    bounds = np.array([[-5.0, 5.0]] * ndim)
+    rng = np.random.default_rng(42)
+    pop_size = 20
+    init_pop = rng.uniform(-5, 5, size=(pop_size, ndim))
+
+    _, _, n_gen = _run_de(
+        cost_fn=trivial_cost,
+        bounds=bounds,
+        init_pop=init_pop,
+        maxiter=500,
+        tol=1e-6,
+        atol=1e-3,
+        mutation=(0.5, 1.0),
+        recombination=0.7,
+        seed=42,
+        n_devices=n_devices,
+    )
+
+    # 简单问题应在远少于 maxiter=500 代内收敛
+    assert n_gen < 500, f"Early stopping 未生效，运行了 {n_gen} 代"
+
+
+def test_run_de_returns_correct_tuple():
+    """验证 _run_de 返回值格式正确。"""
+
+    def const_cost(x: np.ndarray) -> float:
+        return 42.0
+
+    bounds = np.array([[0.0, 1.0]] * 3)
+    init_pop = np.random.default_rng(0).uniform(0, 1, size=(5, 3))
+
+    result = _run_de(
+        cost_fn=const_cost,
+        bounds=bounds,
+        init_pop=init_pop,
+        maxiter=10,
+        tol=1e-6,
+        atol=1e-3,
+        mutation=(0.5, 1.0),
+        recombination=0.7,
+        seed=0,
+        n_devices=1,
+    )
+
+    assert isinstance(result, tuple) and len(result) == 3
+    best_vec, best_cost, n_gen = result
+    assert isinstance(best_vec, np.ndarray)
+    assert best_vec.shape == (3,)
+    assert best_cost == pytest.approx(42.0)
+    assert isinstance(n_gen, int) and n_gen >= 1