STEP: surgical point tracking from sparse endpoints cuts error from 31.5 to 16.8 pixels
ftm_guney · x · 2026-11-03
The STEP paper (arXiv:2610.31951) presents a two-stage framework for surgical point tracking that avoids costly dense trajectory annotations. A frozen offline teacher generates candidate trajectories filtered by temporal reversibility and stereo consistency; a physical endpoint loss and round-trip loss anchor correspondence. On 29 held-out VL-SurgPT clips, endpoint error drops from 31.52 to 16.80 pixels vs dense-label fine-tuning, using a single RGB stream with fixed weights.
More from Embodied
- Yacine's robot now covers 30 square meters in under a minute, target following next — yacineMTB · 2026-11-03
- Marc Schmidt: killing sim2real gap will smash China's actuator moat, up to 60% of humanoid robot cost — MarcJSchmidt · 2026-11-03
- ADAPT: ETH Zurich's 50Hz end-to-end text-driven humanoid control with diffusion priors — breadli428 · 2026-11-03
- Xiaomi open-sources Robotics-1, a robot foundation model trained on 100K hours of trajectories — CyberRobooo · 2026-11-03
- Factory photos show desktop robot modules in volume assembly, project hits production stage — victormustar · 2026-11-03
- Machenta demos instant robot learning from passersby at London King's Cross — Ed__Johns · 2026-11-03