HumanoidBench

HumanoidBench is a challenging whole-body control benchmark used to evaluate the performance of reinforcement learning algorithms in mastering diverse humanoid tasks.

Papers using HumanoidBench (12)

RDA: Reward Design Agent for Reinforcement Learning2026SPHERE: Mitigating the Loss of Spectral Plasticity in Mixture-of-Experts for Deep Reinforcement Learning2026Scaling Tasks, Not Samples: Mastering Humanoid Control through Multi-Task Model-Based Reinforcement Learning2026MVR: Multi-view Video Reward Shaping for Reinforcement Learning2026FastDSAC: Unlocking the Potential of Maximum Entropy RL in High-Dimensional Humanoid Control2026WIMLE: Uncertainty-Aware World Models with IMLE for Sample-Efficient Continuous Control2026Bootstrap Off-policy with World Model2025Flow-Based Policy for Online Reinforcement Learning2025FastTD3: Simple, Fast, and Capable Reinforcement Learning for Humanoid Control2025TDMPBC: Self-Imitative Reinforcement Learning for Humanoid Robot Control2025SimBa: Simplicity Bias for Scaling Up Parameters in Deep Reinforcement Learning2024Rethinking Soft Actor-Critic in High-Dimensional Action Spaces: The Cost of Ignoring Distribution Shift2024