LeRobot Launches Zero-Shot Reward Model API
RemiCadene · x · 2026-07-13
LeRobot v0.6.0 introduces a unified reward models API designed to determine the success of robotic tasks.
It primarily features two zero-shot approaches:
- Robometer: A general-purpose reward model trained on over 1 million trajectories across 21 robot morphologies. By taking video and language instructions from a LeRobot dataset as input, it can directly score task progress and success without needing task-specific retraining.
- TOPReward: Takes it a step further by eliminating the need for any reward weights. It simply wraps an off-the-shelf VLM and reads the log probability of the “True” token under a given video and instruction to assess the outcome.
Additionally, both solutions come with annotation scripts that can write frame-by-frame progress curves back into the dataset, facilitating future reward-aware behavior cloning.
Related event: LeRobot v0.6 Adds Zero-Shot Reward Model API(2 posts)→
More from Embodied
- NVIDIA open-sources FLORA, a robotics data engine ~125x cheaper than GPT-6 Astra — nvidia · 2026-10-27
- Embodied AI startups need a different operational discipline, investor says — ___Mufasaa · 2026-10-27
- π0.5 study: raising demos from 10 to 438 lifts Place Ball in Bowl success from 7% to 83% — stingning · 2026-10-27
- GPT-6 Astra controls dexterous hands zero-shot at just 5.2% success in 500 trials — stingning · 2026-10-27
- HandsOn benchmark: 66 verified tasks and 18,000 human demos for dexterous manipulation — stingning · 2026-10-27
- DexPort: a handheld interface with a real robot hand for dexterous-hand data collection — pabbeel · 2026-10-27