A Model-Free Algorithm for Extremely Resilient Navigation

From Murray Wiki
Revision as of 20:40, 27 December 2023 by Murray (talk | contribs)
(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)
Jump to navigationJump to search
Title A Model-Free Algorithm for Extremely Resilient Navigation
Authors Christian J. Stromberger, Josefine B. Graebener, James F. Ragan and and Richard M. Murray
Source Submitted, 2024 International Conference on Robotics and Automation (ICRA)
Abstract Due to the increasing complexity of space missions and distance to exploration targets, future robotic systems used for space exploration call for more resilience and autonomy. Instead of minimizing the failure risk, we are focusing on missions that will inevitably encounter significant failures and are developing an algorithm that will autonomously reconfigure the system controller to continue to make progress towards the mission goal despite being in a reduced capacity state - we call this extreme resilience. In this paper, we develop a model-free framework to autonomously react to locomotion failures of robotic systems. This is done by the use of a neural network for path planning using the neuroevolution of aug- menting topologies (NEAT) algorithm and a dynamic database of possible moves and their effect on the system’s position and orientation. Two modes of failure detection and resolution are being introduced: (a) relative position failure detection, which is triggered by large, unexpected moves and results in a complete update of the database before a retraining of the neural network, and (b) absolute position failure detection, which triggers from large build-ups of position error from small failures and will induce a retraining of the neural network without an explicit database reset. We implement and validate this framework on a high-fidelity planetary rover simulation using Unreal Engine and on a hardware setup of a TurtleBot2 with a PhantomX Pincher robot arm.
Type Conference submission
URL https://www.cds.caltech.edu/~murray/preprints/str+23-icra s.pdf
ID 2023c