Stanford's RoboArchitect: Coding Model Rewrites Robot Control Code for Under $150, Success Up to 43%
rohanpaul_ai · x · 2026-11-04
A new Stanford paper introduces RoboArchitect. Robot policies often fail on new setups; the usual fix is collecting new demonstrations and finetuning. Wrapping the policy in control code helps, but engineers normally write it by hand for each robot.
Approach: a coding model reviews recorded robot attempts, diagnoses failures, and rewrites the control code over up to 5 rounds — no weight updates, average full-run cost under $150.
Results (all in simulation):
- Unseen-scene success rises from 9.7% (bare policy) to 43.3%;
- Unseen tasks: 12.7% → 40.2%;
- On VLABench, the wrapped general policy beat the same policy finetuned on that benchmark.
More from Embodied
- Sipeed's 10cm MaixStation Dock claims DGX Spark-class local AI inference, runs off a power bank — vincent_koc · 2026-11-04
- D&A turns any VLM into a zero-shot robot reward model, hitting SOTA on 3 benchmarks — KyleMorgenstein · 2026-11-04
- YC shows robots autonomously microwaving edamame with fine-tuned pi0.5 policies — ycombinator · 2026-11-04
- FLoRe teaches humanoid robots force-reactive loco-manipulation from robot-free human demos — HaozhiQ · 2026-11-04
- Dev builds automated wet lab sim with NVIDIA SIMREADY assets for 24/7 robot experiments — davidfromkansas · 2026-11-04
- Direct weight manipulation beats fine-tuning: pi0.5 hits 75% on cloth-folding — ZeeshanZiaML · 2026-11-04