Event
Idealworks and NVIDIA Test COMPASS Navigation Transfer
1 reports · 1 sources
The Idealworks and NVIDIA research teams extended the COMPASS benchmark to conduct sim-to-real evaluation of learned navigation policies at the Idealworks R&D center: they first fine-tuned X-Mobility with digital twin data, then trained COMPASS from scratch in the digital twin, with the policy deployed and running on the robot's onboard Jetson AGX Xavier. Results showed that adding NVIDIA Cosmos-generated sim-to-real data to digital twin training raised real-robot success rates across four scenarios from 65.3% to 91.7%, nearly halved weighted travel time, and narrowed the gap with Idealworks' existing production-line navigation stack to 8.3 percentage points. The evaluation also compared COMPASS trained only with NVIDIA environments against the existing navigation stack, examining task success rate, efficiency, safety, smoothness, and onboard compute. The research team noted that previously there was a lack of systematic testing of whether such adapted policies could remain reliable on real industrial robots; this evaluation also extended to two additional embodiments: a differential-drive robot and an omnidirectional humanoid mobile platform.
Report timeline
- Idealworks and NVIDIA evaluate sim-to-real transfer of COMPASS navigation
Idealworks and NVIDIA systematically evaluated the sim-to-real transfer of the X-Mobility and COMPASS learned navigation pipeline on industrial mobile robots. After training in a digital twin with Cosmos-generated sim-to-real data, real-robot success rates across four scenarios rose from 65.3% to 91.7%, weighted travel time nearly halved, the gap to the production-line navigation stack was 8.3 percentage points, and everything ran on the onboard Jetson AGX Xavier.
arXiv cs.RO ·