Improve Autoware performance on F1/10

Checklist

[X] I've read the contribution guidelines.
[X] I've searched other issues and no duplicate issues were found.
[X] I've agreed with the maintainers that I can plan this task.

Description

General performance

It was discussed in racing WG meeting regarding the general performance of Autoware.Auto on Jetson computers in terms of, for example, FPS and jigger of planning and perception outputs, and the possibility of using an entry level Jetson Nano.

Localization

Current Autoware stack on F1/10 does not feature IMU sensor and KF for odometry, but instead uses odometry generated by motor and steering, unfiltered.

Perception

a 2D object detection package needs to be identified and integrated with Autoware to enable multi-vehicle racing

Planning

for the record replay planner, the recorded trajectories should be generated by a trajectory optimization algorithm.

obstacle collision estimator should be integrated into dynamic trajectory modification.

Control (optional)

Replace pure pursuit with MPC

Operational (optional)

a low-cost, trackside telemetry solution is needed

Certain safety features in planning and control nodes that would estop the vehicle needs to be bypassed to allow longer running time on the track.

Purpose

To meet the need for the upcoming ICRA 2022 F1/10 competition, many can be done

Possible approaches

General Performance

Remove dependency on Rviz and run Jetson in headless mode.

Localization

Configure an IMU as on the original F1/10.

Look into ‘robot_localization’ for KF solutions.

Perception

2D object detection TBD

Planning

Look into IAC teams’ open source trajectory optimization suite

Definition of done

Improve the baseline Autoware.Auto performance on F1/10 hardware in terms of reliability and lap time

autowarefoundation / autoware.universe