Final

[pranavreddy23]

AutoSteer Integration and PathFinder Improvements

Summary

Integrates AutoSteer temporal steering prediction model and improves PathFinder implementation.

Changes

AutoSteer Integration

• Added AutoSteerOnnxEngine for temporal steering angle prediction
• Integrated AutoSteer output into PathFinder as curvature feedforward term
• Added temporal buffer (circular buffer) for EgoLanes raw tensor concatenation
• AutoSteer uses separate TensorRT engine with 1GB workspace
• CSV logging includes autosteer_angle_deg and autosteer_valid columns

PathFinder

• Removed calculateEgoPath() function and center coefficient calculation
• Simplified polynomial fitting to focus on lane detection
• AutoSteer steering angle (radians) now used as curvature input when available

Visualization (visualize_steering.py)

• Added filtering for values outside ±30° range
• Added --trim-start and --trim-end options for frame trimming
• Improved error metrics calculation

Scripts

• Added production-ready run_final.sh with environment variable configuration
• Removed hardcoded paths, improved error handling

Testing

• Verified AutoSteer inference produces valid steering angles
• Confirmed PathFinder correctly uses AutoSteer output
• Tested visualization filtering and trimming

[m-zain-khawaja]

Approved - thanks @pranavreddy23