Hierarchical Off-Road Path Planning and Its Validation Using a Scaled Autonomous Car'

Author

Angshuman Goswami, Clemson University

Date of Award

12-2017

Document Type

Thesis

Degree Name

Master of Science (MS)

Department

Mechanical Engineering

Committee Member

Dr. Ardalan Vahidi, Committee Chair

Committee Member

Dr. John R. Wagner

Committee Member

Dr. Phanindra Tallapragada

Abstract

In the last few years. while a lot of research effort has been spent on autonomous vehicle navigation, primarily focused on on-road vehicles, off-road path planning still presents new challenges. Path planning for an autonomous ground vehicle over a large horizon in an unstructured environment when high-resolution a-priori information is available, is still very much an open problem due to the computations involved. Localization and control of an autonomous vehicle and how the control algorithms interact with the path planner is a complex task. The first part of this research details the development of a path decision support tool for off-road application implementing a novel hierarchical path planning framework and verification in a simulation environment. To mimic real world issues, like communication delay, sensor noise, modeling error, etc., it was important that we validate the framework in a real environment. In the second part of the research, development of a scaled autonomous car as part of a real experimental environment is discussed which provides a compromise between cost as well as implementation complexities compared to a full-scale car. The third part of the research, explains the development of a vehicle-in-loop (VIL) environment with demo examples to illustrate the utility of such a platform. Our proposed path planning algorithm mitigates the challenge of high computational cost to find the optimal path over a large scale high-resolution map. A global path planner runs in a centralized server and uses Dynamic Programming (DP) with coarse information to create an optimal cost grid. A local path planner utilizes Model Predictive Control (MPC), running on-board, using the cost map along with high-resolution information (available via various sensors as well as V2V communication) to generate the local optimal path. Such an approach ensures the MPC follows a global optimal path while being locally optimal. A central server efficiently creates and updates route critical information available via vehicle-to-infrastructure(V2X) communication while using the same to update the prescribed global cost grid. For localization of the scaled car, a three-axis inertial measurement unit (IMU), wheel encoders, a global positioning system (GPS) unit and a mono-camera are mounted. Drift in IMU is one of the major issues which we addressed in this research besides developing a low-level controller which helped in implementing the MPC in a constrained computational environment. Using a camera and tire edge detection algorithm we have developed an online steering angle measurement package as well as a steering angle estimation algorithm to be utilized in case of low computational resources. We wanted to study the impact of connectivity on a fleet of vehicles running in off-road terrain. It is costly as well as time consuming to run all real vehicles. Also some scenarios are difficult to recreate in real but need a simulation environment. So we have developed a vehicle-in-loop (VIL) platform using a VIL simulator, a central server and the real scaled car to combine the advantages of both real and simulation environment. As a demo example to illustrate the utility of VIL platform, we have simulated an animal crossing scenario and analyze how our obstacle avoidance algorithms performs under different conditions. In the future it will help us to analyze the impact of connectivity on platoons moving in off-road terrain. For the vehicle-in-loop environment, we have used JavaScript Object Notation (JSON) data format for information exchange using User Datagram Protocol (UDP) for implementing Vehicle-to-Vehicle (V2V) and MySQL server for Vehicle-to-Infrastructure (V2I) communication.

Recommended Citation

Goswami, Angshuman, "Hierarchical Off-Road Path Planning and Its Validation Using a Scaled Autonomous Car'" (2017). All Theses. 2793.
https://open.clemson.edu/all_theses/2793