Benchmark RadarHumanoidArena
HumanoidArena is a simulation benchmark for egocentric hierarchical whole-body learning, evaluating high-level policies that predict whole-body actions for low-level general motion trackers across seven leg-critical human-object and human-scene interaction tasks, with perturbation-conditioned and GMT-conditioned evaluation.
- Released
- 2026-06-16
- Readiness
- Paper only
- Primary field
- General AI
Why it matters
Existing benchmarks rarely evaluate the policy-tracker interface itself, leaving open whether intermediate whole-body actions are executable, robust under task distribution shifts, and transferable across different GMT backends. HumanoidArena addresses this gap by emphasizing leg-critical interactions and transferable intermediate action representations, providing a basis for comparing hierarchical control architectures.
Motivation
Humanoid robots promise whole-body interaction in human-centered environments, but scalable policy learning remains difficult because task-level decision-making and whole-body dynamic execution are tightly coupled.
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.