Browsing by Author "Warnell, Garrett"

Now showing 1 - 2 of 2
  • Thumbnail Image
    Improving Autonomous Robotic Navigation Using Imitation Learning
    Cèsar-Tondreau, Brian; Warnell, Garrett; Stump, Ethan; Kochersberger, Kevin B.; Waytowich, Nicholas R. (Frontiers Media, 2021-06-01)
    Autonomous navigation to a specified waypoint is traditionally accomplished with a layered stack of global path planning and local motion planning modules that generate feasible and obstacle-free trajectories. While these modules can be modified to meet task-specific constraints and user preferences, current modification procedures require substantial effort on the part of an expert roboticist with a great deal of technical training. In this paper, we simplify this process by inserting a Machine Learning module between the global path planning and local motion planning modules of an off-the shelf navigation stack. This model can be trained with human demonstrations of the preferred navigation behavior, using a training procedure based on Behavioral Cloning, allowing for an intuitive modification of the navigation policy by non-technical users to suit task-specific constraints. We find that our approach can successfully adapt a robot’s navigation behavior to become more like that of a demonstrator. Moreover, for a fixed amount of demonstration data, we find that the proposed technique compares favorably to recent baselines with respect to both navigation success rate and trajectory similarity to the demonstrator.
  • Thumbnail Image
    Towards Fully Autonomous Negative Obstacle Traversal via Imitation Learning Based Control
    César-Tondreau, Brian; Warnell, Garrett; Kochersberger, Kevin; Waytowich, Nicholas R. (MDPI, 2022-06-22)
    Current research in experimental robotics has had a focus on traditional, cost-based, navigation methods. These methods ascribe a value of utility for occupying certain locations in the environment. A path planning algorithm then uses this cost function to compute an optimal path relative to obstacle positions based on proximity, visibility, and work efficiency. However, tuning this function to induce more complex navigation behaviors in the robot is not straightforward. For example, this cost-based scheme tends to be pessimistic when assigning traversal cost to negative obstacles. Its often simpler to ascribe high traversal costs to costmap cells based on elevation. This forces the planning algorithm to plan around uneven terrain rather than exploring techniques that understand if and how to safely traverse through them. In this paper, imitation learning is applied to the task of negative obstacle traversal with Unmanned Ground Vehicles (UGVs). Specifically, this work introduces a novel point cloud-based state representation of the local terrain shape and employs imitation learning to train a reactive motion controller for negative obstacle detection and traversal. This method is compared to a classical motion planner that uses the dynamic window approach (DWA) to assign traversal cost based on the terrain slope local to the robots current pose.