USC's Heave-Ho coordinates humanoid robot teams with zero network messages, 86% success at 120x faster planning
yuewang314 · x · 2026-11-21
USC researchers unveiled Heave-Ho, a decentralized framework for long-horizon multi-humanoid collaboration, with paper, code and project page released.
Key points:
- 120x faster planning with 86% success on long-horizon tasks vs 42% for a GPT-based planner
- Zero inter-robot network messages: a fine-tuned VLM (Qwen3-VL-2B + LoRA) picks team skills; the leader robot speaks commands aloud and followers sync via onboard hearing on a shared schedule
- Training: multi-agent RL post-training (centralized training, decentralized execution) adapts a humanoid foundation tracking model to physically coupled interactions
- Scalability: trained with just 2 identical humanoids, deploys to teams of up to 10 and generalizes to heterogeneous platforms and human partners without retraining
They also released HH-Bench, 12 collaborative task primitives for teams of 2–10 humanoids. The team also showed their G1 'getting a job' at U-Haul.
More from Embodied
- Reverse-engineering Apple Watch: a single Claude Code prompt builds a physics watch face in 30 minutes — mtizard · 2026-11-21
- Robotics Founder: Housekeeping Bots Should Target Mundane Upkeep, Not Party Tricks — Ben_Reinhardt · 2026-11-21
- GPT-6.2 Astra tops AA-Robotics v0.1 with 75% fewer arm commands than Claude — ArtificialAnlys · 2026-11-21
- Heave-Ho: decentralized framework coordinates teams of 2–10 humanoids on long-horizon tasks — yuewang314 · 2026-11-21
- Unitree G1 humanoid robot 'gets a job' at U-Haul in real-world demo — yuewang314 · 2026-11-21
- AgentRoboCritic: action-level benchmark reveals how frontier agents really control robots — XiongChenyan · 2026-11-21