RL²-VLA boosts robot manipulation via RL steering
rsasaki0109 · x · 2026-08-18
RL²-VLA is an adaptive inference-time steering framework that improves robotic manipulation by applying Reinforcement Learning on VLA latents. It trains a lightweight offline RL flow-matching policy and steers the base VLA during inference. The method activates compositional steering mainly when failure is predicted. Benchmarks show average success rate improvements of 10.1% on SIMPLER, 8.9% on PolaRiS, and 19.5% on real robots.
Related event: RL²-VLA Boosts Robot Manipulation via Reinforcement Learning(2 posts)→
More from Embodied
- Humanoid runs full length of indoor track at Beijing Olympic venue — rohanpaul_ai · 2026-08-18
- US-Built Robotics Wins as Tariffs Hit Foreign Supply Chains — MatthewChang · 2026-08-18
- California approves major Waymo expansion into East Bay, Sacramento and rural Sonoma County — moultano · 2026-08-18
- Vision AI Grafting Robot Processes 4,000 Plants Per Hour, Outperforming Skilled Hands — Olivier__OG · 2026-08-18
- Einride brings 500 Tesla Semis to North America, all managed by its Saga AI platform — XFreeze · 2026-08-18
- Booster T2 humanoids rehearse precisely-aligned group locomotion ahead of Beijing's World Humanoid Robot Games — rohanpaul_ai · 2026-08-18