University of Maryland · Embodied AI
UMD tests robot learning, glass perception and visual-tactile memory
What happened
UMD reports that HumanEgo averaged 92.5% success across four robot tasks after 30 minutes of human video per task and transferred to different robots and environments without retraining. PolarDepth improved estimates in real and synthetic glass-walled scenes, while SkillFormer reduced slip-detection delay by 87% in a bottle-capping test.
Why it matters
The results cover specific controlled tasks and workshop demonstrations rather than sustained operation in homes or workplaces. The university report does not establish broad reliability, deployment scale or commercial availability for any of the systems.
Source checked October 31, 2026. Status describes the evidence in the linked source at the time of review.