WEDGE uses navigation agent failures to generate targeted training environments
mohitban47 · x · 2026-10-16
WEDGE (From Failures to Learnable Worlds) introduces policy-conditioned environment generation for embodied navigation agents.
Key findings:
- Even on the same navigation task, different agents fail under different environmental conditions, yet most environment generators ignore what each agent actually needs to practice
- WEDGE discovers environment features as executable functions that explain why an agent fails
- It then generates novel learnable environments targeting those weaknesses, and iterates over multiple rounds to keep finding and fixing new failure modes
More from Embodied
- MIT's C-GMT gives humanoid robots whole-body compliance without sacrificing motion tracking — MIT_CSAIL · 2026-10-16
- OpenRoboto, Bittensor's largest robotics subnet, puts its data collectors on Solana via peaqOS — bittingthembits · 2026-10-16
- Robotics data tokenization landscape: Khala maps peaq, THOR, BitRobot and Axis — 0xSammy · 2026-10-16
- Perseverance rover logs 28.1 miles on Mars, breaking Opportunity's record in half the time — BrettKrieger12 · 2026-10-16
- Trossen bimanual arms portion food in commercial kitchen with models fine-tuned via Physical Intelligence API — kscottz · 2026-10-16
- Safeworld emerges from stealth with $12M to build 'CI/CD for robots' safety testing — lukas_m_ziegler · 2026-10-16