ACCRUE: continual off-policy RL lets robots learn new tasks in 4-12 minutes with 98.3% skill retention
coallaoh · x · 2026-10-15
ACCRUE introduces continual off-policy reinforcement learning for robot foundation models, letting real robots learn new tasks from a small set of demos while retaining prior skills. Built on π0.5 + FastTD3, RL updates only a compact actor-critic while the foundation model stays frozen, keeping GPU overhead low. Actor-critic weights are inherited across tasks with distillation of earlier behavior. On real robots, subsequent tasks reach 95% success in about 4–12 minutes of interaction, robot interaction drops 31.0–68.5% versus single-task RL, and prior-task success holds at 98.3% (118/120 trials across 6 tasks).
Related event: NVIDIA and KAIST Unveil ACCRUE for Continual Real-Robot RL(2 posts)→
More from Embodied
- AlphaNIST: one RL policy solves all 20 NIST assembly tasks with 98.2% success in simulation — hbouammar · 2026-10-15
- Neural field planning for hybrid manufacturing cuts estimated build time 68.3% — jiqizhixin · 2026-10-15
- Work photos are AI's first mile: if capture is wrong, no model can recover — arnabdotorg · 2026-10-15
- LEGO builds an omnidirectional vehicle for automated warehouses — glenbeer · 2026-10-15
- SlipSense: multimodal tactile slip detection hits 96.7% F1 at 23.1ms with zero-shot robot transfer — chris_j_paxton · 2026-10-15
- BlackBerry launches QNX Launchpad to court early-stage Physical AI startups — pdamodaran · 2026-10-15