Occupancy Grid based Model Predictive Control


Mathematically modeling the surrounding environment is an essential part of motion planning for autonomous driving. However, this process becomes very difficult as the environment becomes more irregular and complex. In this regard, “Occupancy Grid” has several advantages. An occupancy grid is a representation of the environment as a grid with occupancy values between 0 and 1, which is very simple. The beauty of occupancy grids is that they can be derived from simply projecting arbitrary spatial information, such as HD map data, object detection results, and even raw data from spatial sensors like LIDAR or 4D Radar. Our research aims to find the optimal local motion plan in complex and uncertain environments by directly applying the spatial information integrated through such occupancy grids to nonlinear model predictive control.



Fig. 1. Examples of the obtained occupancy grid in real time in the same situation (a), utilizing (b) object bounding boxes and road boundary information or (c) LIDAR point clouds. The resulting trajectory plan by at that moment is shown along with the bounding box of the ego vehicle.



Fig. 2. Motion planning tasks with moving objects.



Cho, Minsu, Yeongseok Lee, and Kyung-Soo Kim. "Model Predictive Control of Autonomous Vehicles With Integrated Barriers Using Occupancy Grid Maps." IEEE Robotics and Automation Letters 8.4 (2023): 2006-2013.







GPU Hardware Accelerated Nonlinear Model Predictive Control


Collision avoidance in emergency situations is a crucial and challenging task in motion planning for autonomous vehicles. Especially in the field of optimization-based planning using nonlinear model predictive control, many efforts to achieve real-time performance are still ongoing. Among various approaches, the iterative linear quadratic regulator (iLQR) is known as an efficient means of nonlinear optimization. Additionally, parallel computing architectures, such as GPUs, are more widely applied in autonomous vehicles. In this paper, we propose 1) a parallel computing framework for iLQR with input constraints considering the characteristics of the problem and 2) a proper environmental formulation that can be covered with single-precision floating-point computation of the GPU. The GPU-accelerated framework was tested on a real-time simulation-in-the-loop system using CarMaker and ROS at a 20 Hz sampling rate on a low-performance mobile computer and was compared against the same framework realized with a CPU.


Fig. 3. Algorithm details of GPU-Parallelized iLQR 



Fig. 4. Motion results GPU vs CPU




Lee, Yeongseok, Minsu Cho, and Kyung-Soo Kim. "Gpu-parallelized iterative lqr with input constraints for fast collision avoidance of autonomous vehicles." 2022 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). IEEE, 2022.

MSC Lab.4F(#3434~#3446) ID B/D(N25), School of Mechanical Aerospace & System
Engineering : Division of Mechanical Engineering, KAIST, 373-1, Guseong-dong, Yuseong-gu, Daejoen, Korea, 305-701.
Tel:+82-42-350-3268 l Fax:+82-42-350-5201