TweinStein/OVERTAKING_HANDOFF.md

Handoff: Classical-Control Overtaking Implementation

Context & Goal

We implemented a classical, heuristic-only overtaking state machine for the robot. This feature runs alongside the experimental residual-RL model and executes overtakes by dynamically adjusting the lateral setpoint (dist_ref_cur) of the existing PD wall-follower.

Files Modified

• RTOS_SensorBoard/RTOS_SensorBoard.c

Key Implementations

1. Overtaking State Machine (overtake_state_t):
   • States: S_FOLLOW, S_DETECT, S_COMMIT, S_PASS, S_REJOIN.
   • Modulates dist_ref_cur to smoothly shift the robot out of line and past an opponent without needing a new low-level controller.
2. Dynamic Object Classifier:
   • Uses an ego-motion-compensated range-rate classifier to differentiate between static walls (which approach at ego speed) and moving cars.
   • Calculates a residual (r = front_filt - front_pred) and accumulates it (r_acc).
3. Model Gating:
   • Model_ApplyResidual(&throttle_l, &throttle_r, &steeringAngle) is now strictly gated to only apply when ot_state == S_FOLLOW. It won't interfere with the passing logic.
4. Logging (FileDumpRow):
   • Increased NUMCOLS to 14.
   • Now logs ot_state and dist_ref_cur alongside standard telemetry for post-run analysis.
5. Calibration Function (RobotCalib):
   • Added a dedicated routine to calibrate the throttle-to-velocity mapping (V_MAX_MMPS).
   • Drives perfectly straight (steeringAngle = 3100;) at CALIB_THROTTLE.
   • Calculates -d(front)/dt over the period where 50 <= front <= 1000 (after a 500ms startup delay) and outputs the average velocity to the ST7735 display.

Current Status

• The code is fully integrated into RTOS_SensorBoard.c but requires empirical testing and tuning.

Next Steps for the Next Agent/User

1. Run RobotCalib(): Hook up RobotCalib to S2Push() or run it directly. Let the robot drive at a wall/target and observe the reported "Avg Vel".
2. Update Parameters: Use the calibration result to update V_MAX_MMPS (currently 2000 as a placeholder) at the top of RTOS_SensorBoard.c.
3. Bench Testing: Verify the state machine logic transitions correctly when waiving an obstacle in front of the sensors.
4. Solo On-Track Tuning: Drive the robot alone on the track. If the classifier fires (transitions to DETECT or COMMIT) on apexes/corners, tune R_YAW_GAIN_Q8 and R_THRESH_BASE until the false positives disappear.
5. Opponent Testing: Test passing on straights using stationary and moving opponents. Watch for edge/corner cases or premature rejoins (T_PASS_MIN, T_CLEAR_MS).