Predicting Positions of People in Human-Robot Conversational Groups
summary
System builds heat maps of where people stand relative to an anchor participant by aligning formations with transformation matrices, trains a classifier on SALSA and Babel (80/20 train/test; Babel = 35 min, 7 people), and implements two models to estimate true group size from partial detections and to predict locations of undetected participants. Evaluation uses confusion matrices with strong diagonals, slightly lower accuracy on noisy SALSA, and on-robot demos succeed at detecting missing members and turning the robot toward predicted positions.
AI-written summary · about
related talks
09mSituated Live Programming for Human-Robot Collaboration
05mHuman-in-the-loop Pose Estimation via Shared Autonomy
28mPreference Learning from Minimal Human Feedback for Interactive
15mSensing and Handling Engagement Dynamics in Human-Robot Interaction Involving Peripheral ...
16mRobotic Telepresence at Scale
21mVisualizing robot minds, and other fun stuff
21mDistributed Reinforcement Learning for Robotic Assembly
13mEnhancing Object Detection & Avoidance Systems for Robots in Challenging Environments Using AI