Hebero Benchmark for Heterogeneous Multi-Task Robot Learning
October 5, 2026
Hebero is a GPU-parallel Isaac Lab benchmark designed for training single policies across 40 different manipulation tasks. It introduces Demonstration-Guided Policy Optimization (DGPO) to handle sparse rewards by reusing demonstrations for dense tracking rewards.
HOW THIS AFFECTS YOU
●
researcherYou can evaluate multi-task RL policies more efficiently using GPU-parallelized heterogeneous environments.