CoRL researcher unpacks "agents are too slow for robots" into three distinct critiques
tomssilver · x · 2026-11-11
At CoRL 2026, robotics researcher Tom Silver argues that the common criticism "agents may be effective at robot decision-making but are too slow / not reactive" (often used to motivate VLAs and WAMs) actually conflates three very different issues:
- (a) Slow per step: the agent sits in the control loop, queried for every action — a genuine latency problem.
- (b) Slow per task: the agent plans or writes code for each new task, then the robot executes — an application-dependent tradeoff, not a flaw.
- (c) Slow per domain: the agent writes a policy once, offline, and isn't used at deployment — an amortized cost competing with engineering or training time.
He adds that the related claim "agents aren't closed-loop / reactive" only holds for (a): in (b) and (c), the agent outputs a program, and a program can itself be a closed-loop policy running at control rates. He calls for better terminology to distinguish these cases.
More from Embodied
- Baidu World 2026 to showcase Apollo Go robotaxi rides and Apollo World physical AI model — Baidu_Inc · 2026-11-12
- NVIDIA's open-source Alpamayo 2 Edge VLA driving model lands in Autoware — drmapavone · 2026-11-12
- A closer look at Astra's robot gripper design (and a joke at its expense) — _Stocko_ · 2026-11-12
- Robot startup passes first presale milestone with 10+ units claimed — R4nsack · 2026-11-12
- Runway launches Praxis-1, a 27B MoE world action model, with a 73ms Fast variant — ChongZzZhang · 2026-11-12
- Dream-1 on bimanual YAM robot folds a towel on voice command at CoRL — DJiafei · 2026-11-12