- 68 views
- 384 downloads
Optimal Motion Planning in GPS-Denied Environments using Nonlinear Model Predictive Horizon
-
- Author / Creator
- Al-Younes, Younes M. A.
-
Navigating robotic systems autonomously through unknown, dynamic, and GPS-denied environments is a challenging task. One requirement of this is a path planner which provides safe trajectories under real-world conditions such as nonlinear vehicle dynamics, real-time computation requirements, complex 3D environments, and moving obstacles. The limitations of existing methods for path planning related to computational efficiency, consideration of complex system dynamics, and achieving consistency and optimality of the solution, have pushed our research towards proposing an approach that tackles these challenges all at once.
The goals of the proposed research are threefold. The first goal is proposing a novel framework for trajectory generation based on optimization and implementing it on a multi-rotor drone. The nonlinear dynamics of a drone make it an excellent test candidate for this work. The proposed trajectory planning system, named Nonlinear Model Predictive Horizon (NMPH), extends the well-known Nonlinear Model Predictive Control (NMPC) methodology by combining it with a nonlinear control design, for instance FeedBack Linearization (FBL) or BackStepping Control (BSC). The purpose of embedding a nonlinear control law within the optimization is to reduce the non-convexity of the problem and thus provide faster solutions for the trajectory planning problem. NMPH provides feasible solutions, generates smooth and collision-free trajectories, supports moving obstacles, is able to run in real-time, and reduces battery power consumption by producing smooth flight trajectories for the drone to follow.
The second goal is developing a global motion planning system to allow the drone to explore a complex and GPS-denied environment with the presence of both static and dynamic obstacles. This system is a three-stage modular design that incorporates the NMPH trajectory planning algorithm and a graph-based planner. The first stage operates by building an incremental map of the environment, also containing a volumetric representation of the static and dynamic obstacles. The second stage is a graph-based planner that generates waypoints within unexplored areas of the map. The final stage uses the NMPH algorithm to produce continuous and optimal trajectories from the vehicle's current pose to the waypoints generated by the graph-based planner. For a smooth integration between the three layers, computationally efficient algorithms for obstacle mapping and avoidance plus robust path guidance algorithms are developed. The presented approaches are then implemented in software to generate optimal paths for a drone navigating an unexplored GPS-denied environment, with several simulations and experimental results provided to demonstrate the features and evaluate the performance of the overall design.
The third goal is proposing an adaptive learning scheme for the NMPH algorithm based on Deep Reinforcement Learning (DRL). The resulting design is called `adaptive NMPH', which generates optimal trajectories for an autonomous drone based on the system's states and its environment. This is done by online tuning the NMPH's optimization parameters using two different Actor-Critic DRL-based algorithms, Deep Deterministic Policy Gradient (DDPG) and Soft Actor-Critic (SAC). Both adaptive NMPH variants are trained and evaluated on an aerial drone inside a high-fidelity simulation environment. The results demonstrate the learning curves, sample complexity, and stability of the DRL-based adaptation scheme, and show the superior performance of adaptive NMPH relative to non-adaptive designs. -
- Graduation date
- Spring 2023
-
- Type of Item
- Thesis
-
- Degree
- Doctor of Philosophy
-
- License
- This thesis is made available by the University of Alberta Libraries with permission of the copyright owner solely for non-commercial purposes. This thesis, or any portion thereof, may not otherwise be copied or reproduced without the written consent of the copyright owner, except to the extent permitted by Canadian copyright law.