USC lab's SIMPLE lands at CoRL 2026: 60-task humanoid sim benchmark for VLA models
zhengyiluo · x · 2026-09-08
- What it is: SIMPLE, a simulation-based policy learning and evaluation testbed for humanoid loco-manipulation from USC's PSI Lab, accepted to CoRL 2026 with paper, code, and project page public.
- Stack: MuJoCo contact-rich dynamics coupled with Isaac Sim photorealistic rendering; 60 whole-body tasks across 50 indoor scenes with 1,000+ object assets.
- Data pipelines: automated trajectory generation via motion planning plus a low-latency VR teleoperation interface, claiming up to 3× faster data collection.
- Benchmark: unified parallel rollouts and leaderboard for mainstream policies incl. GR00T, DreamZero, Cosmos3, π₀, π₀.₅.
- Key finding: sim performance strongly correlates with the real world, and policies trained on SIMPLE data transfer zero-shot to physical humanoids.
More from Embodied
- EmbodiedSkills: A Unified Framework for Orchestrating VLA Robot Agents — Wei Wang · 2026-09-08
- Dev Builds a TikTok-Style Mobile UI for 15+ Years of Replication Archives — Josikinz · 2026-09-08
- Swarm Robotics Paper Tests When Sociality Pays Off: Scarce Environments Erase the Advantage — abenitezburraco · 2026-09-08
- GPT-6 Astra builds 3D site exploding a humanoid robot into 1,168 CAD parts — freelerobot · 2026-09-08
- MotionVLA accepted at CoRL 2026: remember motion, not frames — yshan2u · 2026-09-08
- Robotics researcher: harness + tool calls beat raw VLA on control benchmarks, at 2.3x lower cost — rbhar90 · 2026-09-08