Corridor-aware planner: boundary tracking, trap-avoidance, race-line heuristic, tests & Docker

[Weiores]

Motivation

• The original planner selected the deepest LiDAR gap and was vulnerable to trap openings and off-track choices, so a corridor-aware approach was needed to respect track boundaries and avoid unsafe openings.
• The vehicle should follow a more optimal racing line (outside-inside-outside) rather than simply aiming for the largest open bearing.
• Changes must be limited to the gap_finder package and be verifiable with pure-Python unit tests and a Docker-based test runner.

Description

• Add a new CorridorPlanner in gap_finder/gap_finder/planner.py that sanitizes scans, extracts left/right boundary features with confidence, computes a safe corridor, detects trap openings, applies a race-line bias, and scores candidate angles (clearance + race-term + smoothness) rather than taking a raw max-range index.
• Replace the old logic in scripts/gap_finder_base.py to call the new planner and expose tunable parameters (e.g. track_width_estimate, boundary_confidence_threshold, trap_range_jump_threshold, race_line_bias_strength, steering_smoothing_alpha, speed_reduction_on_low_confidence).
• Add unit tests f1tenth_ws/src/gap_finder/tests/test_planner.py that use synthetic LaserScan-like arrays for straight corridor, left curve, trap opening, and missing-boundary scenarios.
• Add f1tenth_ws/src/gap_finder/Dockerfile and update README.md with exact commands to build/run tests in Docker, and export planner types via gap_finder/gap_finder/__init__.py.

Testing

• Ran the unit test suite locally with PYTHONPATH=f1tenth_ws/src/gap_finder pytest -q f1tenth_ws/src/gap_finder/tests which completed successfully: 4 passed.
• Attempted to run the Docker build from the environment with docker build -t gap-finder-tests -f f1tenth_ws/src/gap_finder/Dockerfile f1tenth_ws/src/gap_finder but the docker CLI is not available in this environment, so the Docker build/run could not be executed here.

---

Codex Task