Research map / Multi-agent systems and games / Human-AI and goal-conditioned
CoMaTrack: Competitive Multi-Agent Game-Theoretic Tracking with Vision-Language-Action Models
Li Gao, Liu Liu, Mingyang Lyu, Yang Cai
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add single-agent GRPO RL fine-tuning after SFT (ablation) | CoMaTrack (Qwen2.5VL-3B) · EVT-Bench STT | SR / TR / CR: 88.2 -> 89.5 SR; 85.4 -> 88.0 TR; 3.1 -> 2.2 CR | Improved |
| use competitive multi-agent RL instead of single-agent RL (ablation) | CoMaTrack (Qwen2.5VL-3B) · EVT-Bench STT | SR / TR / CR: 89.5 -> 92.1 SR; 88.0 -> 90.3 TR; 2.2 -> 0.9 CR | Improved |
| add competitive multi-agent RL on top of SFT (ablation) | CoMaTrack (Qwen2.5VL-3B) · EVT-Bench STT | SR: 88.2 -> 92.1 | Improved |
| use a static-obstacle opponent during multi-agent RL (ablation) | CoMaTrack · CoMaTrack-Bench | SR vs training episodes | No clear effect |
| use a random-interference opponent during multi-agent RL (ablation) | CoMaTrack · CoMaTrack-Bench | SR vs training episodes | Hurt |
| use a competitive-tracking opponent (SFT policy) during multi-agent RL (ablation) | CoMaTrack · CoMaTrack-Bench | SR vs training episodes | Improved |
| train 3B CoMaTrack with multi-agent RL versus 7B TrackVLA++ | CoMaTrack vs TrackVLA++ · EVT-Bench STT | SR: 90.9 -> 92.1 | Improved |
| train 3B CoMaTrack with multi-agent RL versus 7B TrackVLA++ | CoMaTrack vs TrackVLA++ · EVT-Bench DT | SR: 74.0 -> 74.2 | Improved |
| train 3B CoMaTrack with multi-agent RL versus 7B TrackVLA++ | CoMaTrack vs TrackVLA++ · EVT-Bench AT | SR: 55.9 -> 57.5 | Improved |
| CoMaTrack versus Uni-Navid zero-shot | CoMaTrack vs Uni-Navid · CoMaTrack-Bench | SR / TR / CR: 42.4 -> 85.0 SR; 56.5 -> 82.9 TR; 23.8 -> 5.5 CR | Improved |
Closest papers by meaning
- Long-term Embodied Visual Tracking with Lightweight Vision-Language-Action Models (NeurIPS 2026 · similarity 0.88)
- Mimicking To Dominate: Imitation Learning Strategies for Success in Multiagent Games (NeurIPS 2024 · similarity 0.86)
- GenRL: Multimodal-foundation world models for generalization in embodied agents (NeurIPS 2024 · similarity 0.85)
- VIKI‑R: Coordinating Embodied Multi-Agent Cooperation via Reinforcement Learning (NeurIPS 2025 · similarity 0.85)
- Deep Demonstration Tracing: Learning Generalizable Imitator Policy for Runtime Imitation from a Single Demonstration (ICML 2024 · similarity 0.85)
- Diffusing States and Matching Scores: A New Framework for Imitation Learning (ICLR 2025 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
