NVIDIA releases Hebero: GPU-parallel multi-task robot RL benchmark spanning 40 tasks
nvidia · hf · 2026-10-10
NVIDIA released Hebero, a GPU-parallel Isaac Lab benchmark for heterogeneous multi-task robot RL, jointly training one policy across 40 manipulation tasks.
- More parallel replicas per task improve success under fixed wall-clock budget
- DGPO reuses demonstrations for dense tracking rewards and asymmetric value learning
- IW-ABC coordinates adaptive behavior cloning with importance weighting via per-task learning progress
- With 50 demos per task, IW-ABC hits 90.1% mean success (state inputs), beating FAMO-ABC by 7.8 points; the visual variant reaches 93.5%
- A single sim-trained policy completes 4 tasks on a real Piper robot
Project page: hebero-rl.github.io
More from Embodied
- Physical attention bias cuts cable-simulation prediction error by 15%+ — Avihai Giuili · 2026-10-10
- Wayve's AI Driver hits Turin roads in a Maserati Grecale just 4 weeks after integration — alexgkendall · 2026-10-10
- IROS 2026 Best Paper: humanoid robot learns tennis rallies from 5 hours of human motion data — qinzytech · 2026-10-10
- Tesla Teases Robotaxi "Floodgates Are Opening" as Expansion Signals Build — julianibarz · 2026-10-10
- Adaptive Sampling + Pure RL Yields Robot Policies Beyond Prior Locomotion Limits — jeffclune · 2026-10-10
- Atomic Machines Emerges from Six Years of Stealth with AI-native Matter Compiler that Builds Micro-machines from Code — Merzmensch · 2026-10-10