1 of 33

Group 3:

Overtaking Planner for AI Racing

Vehicle in ROAR

2 of 33

Running over the Front Vehicle Is always a Way to Overtake

Overtaking Planner for AI Racing

Vehicle in ROAR

3 of 33

Running over the Front Vehicle Is always a Way to Overtake

Overtaking Planner for AI Racing

Vehicle in ROAR

4 of 33

Sections

Group Introduction

Background

Algorithm

Implementation

Test Results

Following-up Suggestions

5 of 33

Who We Are

Chenyan Zhang M. Eng. (EECS)
Yunhao Liu M. Eng. (EECS)
Ray Xi B. Sc. (EECS)

6 of 33

Why We Need a Planner for ROAR

Because someone must be responsible for crashing

7 of 33

Why We Need a Planner for ROAR

Because someone must be responsible for crashing

Because the regulations ask us to:

Following mode
Overtaking mode

8 of 33

Why We Need a Planner for ROAR

Because someone must be responsible for crashing

Because the regulations ask us to:

Following mode

Tracing the optimal trajectory
As fast as possible
But never passing or crashing

Overtaking mode

9 of 33

Why We Need a Planner for ROAR

Because someone must be responsible for crashing

Because the regulations ask us to:

Following mode

Tracing the optimal trajectory
As fast as possible
But never passing or crashing

Overtaking mode

Safety-first
Find a path to overtake

10 of 33

How the Planner Work?

Instead of running over the front vehicle

as we all expect (no way)

11 of 33

How the Planner Work?

Derived based on prior work by

He et al. [1]
Zeng et al. [2]

12 of 33

How the Planner Work?

Baseline: MPC

That easy?

13 of 33

How the Planner Work?

Solution: Learning-based MPC

14 of 33

How the Planner Work?

Solution: Learning-based MPC

15 of 33

How the Planner Work?

16 of 33

How the Planner Work?

How to Overtake?

Speed-up and run them over!

17 of 33

How the Planner Work?

How to Overtake?

Speed-up and run them over!

Bezier-curve based planner

+ Learning-based MPC

+ CBF

18 of 33

How the Planner Work?

Bezier-curve based planner

+ Learning-based MPC

+ CBF

19 of 33

How the Planner Work?

Bezier-curve based planner

+ Learning-based MPC

+ CBF

20 of 33

How the Planner Work?

Bezier-curve based planner

+ Learning-based MPC

+ CBF

21 of 33

Coding: Ctrl-C Ctrl-V

No of course.

22 of 33

Coding: What We Did

Refactoring codes

Simplify structures
Documentation!

23 of 33

Coding: What We Did

Refactoring codes

Simplify structures
Documentation!

Re-build Interfaces

Simulator
ROS1 to ROS2

24 of 33

Coding: What We Did

Refactoring codes

Simplify structures
Documentation!

Re-build Interfaces

Simulator
ROS1 to ROS2

Add following mode

MPC Constraints
Fine-tuning

25 of 33

Coding: What We Did

Refactoring codes

Simplify structures
Documentation!

Re-build Interfaces

Simulator
ROS1 to ROS2

Add following mode

MPC Constraints
Fine-tuning

Hardware deployment

Impossible

26 of 33

Coding: What We Did

Refactoring codes

Simplify structures
Documentation!

Re-build Interfaces

Simulator
ROS1 to ROS2

Add following mode

MPC Constraints
Fine-tuning

Hardware connection interface

Enabling deployment
Via ROS2

27 of 33

Test results

Let’s see crashing!

28 of 33

Test results

Following in the outer lane

29 of 33

Test results

Following in the inner lane

30 of 33

Test results

Even crazy track!

31 of 33

Anyone interested in Taking over?

Here are some suggestions!

Linearization & Regression: singularity
Lower-level emergency brake
Reinforcement learning or decision tree

as higher-level planner

We Prepared Documentation for You!

32 of 33

Reference

[1] S. He, J. Zeng, and K. Sreenath, ‘Autonomous racing with multiple vehicles using a parallelized optimization with safety guarantee using control barrier functions’, in 2022 IEEE International Conference on Robotics and Automation (ICRA), 2022.

[2] J. Zeng, B. Zhang, and K. Sreenath, ‘Safety-critical model predictive control with discrete-time control barrier function’, in 2021 American Control Conference (ACC), 2021, pp. 3882–3889.

33 of 33

Any Questions?

Try clicking those to jump back 😉