Contrastive RL over action chunks yields +31.7% offline and +93.1% online gains
burny_tech · x · 2026-09-04
Researchers extended self-supervised contrastive RL from single actions to action chunks, finding large gains: +31.7% offline and +93.1% online.
While many explanations exist for why chunking helps, this work offers a new one: chunking improves the critic's learned representations — a practical insight for RL training in robotics and sequential decision-making.
Related event: Action Chunking Boosts Contrastive RL by 93% in Online Settings(3 posts)→
More from Embodied
- Open multilingual ASR dark horse: Hojo-ASR-Multi-V1 hits 3.54% WER, ranks #7 globally — rohanpaul_ai · 2026-09-04
- Yacine: Precisely modeling a single actuator is a waste — randomize parameters with an RNN instead — yacineMTB · 2026-09-04
- Oura files for US IPO under ticker OURA after selling 3.6M+ smart rings — Polymarket · 2026-09-04
- Blogger applies to buy 25+ Tesla Cybercabs, plans fully public Robotaxi fleet journey — lasas · 2026-09-04
- Editing video with Fable 5.1: a sneak peek at the Daso kids' AI computer — NERDDISCO · 2026-09-04
- Musk confirms Cybercab UI is built on Unreal Engine, crediting Epic Games — elonmusk · 2026-09-04