PolicyTrim cuts VLA robot task time by making action chunks longer and trajectories shorter
新智元 · wechat · 2026-07-21
A new paper proposes **PolicyTrim**, a post-training method for VLA robots that improves real-world execution efficiency without changing the architecture or retraining from scratch. - The core problem is not just inference latency, but **policy efficiency**: how many reliable action chunks and physical steps are needed to finish a task. - PolicyTrim uses a two-stage RL-style post-training pipeline: - **Reliable Action Chunk Extension**: expands the part of an action chunk that can be safely executed. - **Redundancy-Aware Step Reduction**: rewards shorter successful trajectories and suppresses wasted corrective motions. - Reported gains include **3× action-chunk utilization**, up to **51.4% fewer physical steps**, and up to **5.83× end-to-end speedup** on LIBERO with π0.5, while keeping success rate above **98%**. - The method was also tested on **ManiSkill**, **Meta-World**, and real robot tasks on an **Agilex Piper** arm, where it delivered about **1.86×** speedup in the standard setup. - The paper’s key claim is that making robots “take fewer wrong steps” is as important as making each inference faster, and the two approaches can stack.
More from Embodied
- Creator buys a Reachy robot and asks what to build first — dee_hw · 2026-07-21
- MW team shows Gen1 of MW-bot, a semi-humanoid home robot built for pantry storage and ceiling rails — CyberRobooo · 2026-07-21
- Kunlun Spins Out Riemann Dynamics, Unveils World Action Model for Robotics — 0xAllen_ · 2026-07-21
- Gritt exits stealth with $34M to automate construction sites, starting with solar plants — TechCrunch AI · 2026-07-21
- BrainCo demos near-real-time bionics without implants and claims 85% lower prosthetic cost — TrueOrange9944 · 2026-07-21
- OpenAI’s $230 CodexMicro sold out, and users are already cloning it with Stream Decks — APPSO · 2026-07-21