Temporal assessment of excavator proficiency
Multi-agent assessment across 119 operators and three skill levels.
Measure. Interpret. Coach.
My doctoral research asks how recorded behavior can support useful, evidence-grounded feedback after a complete simulator session.
Three research stages
Stage 1 Manuscript
Measure behavior
Analyzed joystick-mounted inertial measurements and separate gaze measures to study experience-related behavior.
94.5% mean balanced accuracy
Stage 2 Planned study
Validate AI interpretation
Design an evaluation of phase identification, behavioral interpretation, and propagation of upstream errors.
Stage 3 Planned study
Evaluate evidence-grounded coaching
Design a session-end workflow for behavioral analysis, reference retrieval, feedback drafting, and verification.
Behavioral measurement has manuscript results. Interpretation and coaching remain planned studies. Learning improvements have not been established.
Related research and infrastructure
Zero-shot locomotion prediction from vision and speech
87.87% weighted F1 from GPT-4o without task-specific training.
Context-aware locomotion prediction
94.3% reported accuracy with local perception, API reasoning, and a clarification gate.
Multimodal research data infrastructure
Supporting data organization for synchronized operator-behavior research.
The wearable-sensor analysis uses N=36 usable IMU trials and a separate N=40 gaze sample. The assessment study covers N=119 operators. These are separate evidence cohorts.
Read the research records