M1 — V0: Interface (July, ~20h)

• Skeleton for Hungarian association module (interface only)
• Skeleton for Kalman fusion module (interface only)
• CARLA simulation with 2+ smart data agents (ego + ≥1 collaborator), exercising the module end-to-end with placeholder messages

Deliverable: Interface spec + skeleton algorithm modules + working multi-agent scenario

M2 — V1: Fully Functional Model (August, ~25h)

• Hungarian algorithm: cost matrix construction + assignment solver, unmatched track/detection handling
• Kalman fusion tracker: predict/update steps, track lifecycle (init/coast/delete)
• Full integration into CARLA multi-agent scenario

Deliverable: Working collaborative perception pipeline — CAM/CPM-like messages in, fused tracks out

M3 — Testing & Validation (September, ~35h)

• Accuracy comparison vs. ground truth (position/velocity error, ID switches, false positives/negatives)
• Edge cases: message loss, latency, sparse CPM coverage
• Bug fixing / tuning (Kalman Q/R, Hungarian cost thresholds)

Deliverable: Validation report + test suite