Human-in-the-loop Pose Estimation via Shared Autonomy
summary
A mixed-initiative pose estimator combines a human click-and-drag initialization with a sampling-based machine step that compares rendered hypotheses to the depth image and converges over time, claiming lower human time for the same error and similar final accuracy. The method uses snap to grid inspired by Monte Carlo localization, runs machine updates every 0.5 s, and delegates translation to the algorithm while leaving rotation to humans. Evaluation uses a Fitts‑law–style analysis and a user study with 8 novice and 8 expert users and motion execution via Affordance Template.
AI-written summary · about
chapters (5)
related talks
28mPreference Learning from Minimal Human Feedback for Interactive
10mEnabling Robot-assisted Motion Capture with Human Scale Tracking Optimization
10mPredicting Positions of People in Human-Robot Conversational Groups
31mData as an instrument for more efficient collaboration with robots
59mOn Human-robot Interactions via Multimodal Intent Modeling and Reachability-based Safety Assurance
14mTowards Unobtrusive Physical AI: Augmenting Everyday Objects with Intelligence and Robotic Moveme...
14mHuman-in-the-Loop Reinforcement Learning
10mDetecting 3D Hand Pointing Direction from RGB-D Data in Wide-Ranging HRI Scenarios