Imitation Learning Overview
Study scope, what the earlier controller achieved, and what it could not represent.
docs_note/imitation_learning.md
Open note →
shared archive
Research Notebook
Technical notes behind the Go1 navigation studies.
This page collects the supporting notes as one shared archive. The markdown files stay as the source of truth, while each note opens through the site note viewer rather than as a raw document.
Phase 1 — Motion Learning
Notes from the imitation-learning navigation study: learning motion behavior, telemetry signals, and real-time deployment boundaries.
Model & Pipeline
ResNet backbone, temporal context, and deployment-oriented motion-stack choices.
docs_note/model.md
Open note →
Data & Telemetry
ROS bags, demonstrations, and low-level supervision used in the motion study.
docs_note/data.md
Open note →
Experiments & Evaluation
Phase 1 evaluation context and the questions that pushed the project toward social decisions.
docs_note/experiments.md
Open note →
Deployment Notes
Real-time boundary notes for what actually ran on Go1 in the earlier controller stack.
docs_note/deployment.md
Open note →
Phase 2 — Social Navigation
Notes from the VLM-based social-navigation study: decision labels, benchmark design, online integration boundaries, observations, and deployment boundary.
VLM Social Navigation Overview
Study scope, benchmark framing, and how the later phase combines offline reporting with online integration experiments.
docs_note/vlm_social_navigation.md
Open note →
Decision Representation
Five-label action space, prompt-policy structure, and semantic decision framing.
docs_note/model.md
Open note →
Benchmark Design
Curated bags, sequence windows, and the saved-run evaluation setup.
docs_note/experiments.md
Open note →
Observations
Where sequence helped, where activation failed, and why uncertainty still mattered.
docs_note/experiments.md
Open note →
Deployment Boundary
Why the VLM stayed advisory and how safety projection fits the shared system view.
docs_note/deployment.md
Open note →
Future Direction
Open questions and next steps connecting motion learning, semantic reasoning, and safety layers.
Future Work
Open questions around arbitration, REVIEW calibration, crossing robustness, and future controller integration.
docs_note/future_work.md
Open note →
Research Artifacts
Compact project artifacts connected to the report, slides, and future technical directions.
Final Presentation
Visitor-facing PDF export of the final VLM social-navigation presentation.
assets/vlm-final-report.pdf
Final Presentation (PDF) →
Future Work Notes
Engineering roadmap note connecting motion learning, semantic reasoning, and safety projection.
docs_note/future_work.md
Open note →
Benchmark Notes
Saved-run setup, metrics, and failure-case interpretation behind the VLM evaluation.
docs_note/experiments.md
Open note →