Status Update: 2022-11-02

[heyong4725]

Dora-Rover

• Control the car is now ok. We have succeeded in overriding the control and are currently using simulated radio control (rc).
• Changed the planner from hybrid_astar to frenet. This makes the waypoints a lot more stable at the cost of a more complex planner. The stability is really useful as the localization is noisier in real life.
• Matched the image with the LIDAR point cloud to enable Obstacle Detection
• Going to move away from LIDAR after this preliminary experiment.
• Using AGX Xavier, dora-rover is using too much CPU/GPU and we're seeing slow down. We're going to upgrade the CPU/GPU card and see if it is better.
  -> Intel x86, industry PC
  -> Nvidia AGX Orin ARM CPU/GPU
  -> Qualcomm CPU/GPU CUDA Compatible

LIDAR?

• Two school of thoughts:
  • Car should be able to do vision only car using BEV. Vision only solution.
  • Vision + LIDAR + ...
  • Privacy concern place is better with LIDAR.

Complete Pylot dataflow

Completely python dataflow on simulation. Soft Real Time. Deadline. Watermark. Pylot.

AutoCore Meeting

Pure C++ Codebase port over dora.
Remove all the dependency.
Features they need:
- request-reply pattern

Small iteration of porting Autoware.

Add README documentation