SimEX lets coding agents learn robot skills in sim with ~10 min of real robot time
micoolcho · x · 2026-11-04
Robo Papers teases an interview with the authors of SimEX (Simulation-Integrated Robotics AutoResearch). SimEX uses simulation as a laboratory where coding agents discover and refine reusable robot skills, then adapt them to real hardware — no demonstrations, no hand-written skills, 10 minutes of robot time per task.
Two stages
- Stage 1: a probe-and-optimize harness drives open-ended cycles in a simulated sandbox — the agent probes its toolbox on self-invented tasks, turns failures into sub-goals, and optimizes until each sub-goal passes.
- Stage 2: the toolbox is adapted to the real robot with only five physical trials, replaying each in simulation to diagnose failures and screening candidate repairs before the next trial.
The project page shows real dual-arm YAM rollouts (barcode scanning etc., at 8x speed) plus the generated toolbox and CaP scripts.
More from Embodied
- TouchBridge trains dexterous robots from human-only demos, hitting 72.5% success rate — ZeYanjie · 2026-11-05
- NYU team open-sources MABEL, an $8K mobile bimanual robot built in 2 weeks — RemiCadene · 2026-11-05
- Turning an AirPods case into a working mouse — TheMoonMidas · 2026-11-05
- Tsinghua, NYU and Northwestern unveil RoboDoctor, a training-free framework for fixing robot actions — jiqizhixin · 2026-11-05
- Robotics exec says deployments start in Singapore, aiming for everywhere from cafes to factories — micoolcho · 2026-11-05
- Gus Robotics watches its Microduck robot production line run for the first time in Shenzhen — mishig25 · 2026-11-05