E^2-CARE: New framework for safe, context-aware caregiving robots
Cornell researchers’ E^2-CARE adapts robot skills zero-shot across diverse environments and embodiments.
Deep Dive
Researchers propose E^2-CARE, a framework for physical caregiving robots that adapts skills in real time to different environments, robot embodiments, and human contexts. It represents primitive skills as interaction templates reshaped online via a unified 3D dynamic scene graph. Tested across hundreds of simulated household environments and real-world settings with two robots on four activities of daily living, E^2-CARE achieved zero-shot skill reuse and demonstrated consistent, safe adaptation.
Key Points
- E^2-CARE uses a unified 3D dynamic scene graph to model robot, environment, and human interaction contexts explicitly.
- Achieved zero-shot adaptation across hundreds of simulated home environments and real-world tests with two different robot embodiments.
- Validated on four activities of daily living (e.g., feeding, dressing) with user studies confirming safe, consistent performance.
Why It Matters
Enables caregiving robots to safely adapt to any home and robot type, accelerating real-world deployment of assistive robotics.