Carleton University
Bio-Inspired Adaptive Control of Robotic Manipulators for Grasping in Orbital Space Missions
Abstract
dc:description.abstractThis research addresses the control problem of spaceborne manipulators for adaptive and reactive grappling of space targets for orbital missions requiring faster dynamics interactions than are currently employed. One of the major difficulties in validation of space manipulator algorithms is the challenge and expense of experimentally replicating the space environment on earth. This research’s approach to solving this problem has evolved from a bio-inspired feedforward approach to a proprioceptive reinforcement learning approach. Initially it was assumed that a bio-inspired feedforward approach could provide human-like tactility required for robustness and adaptability in space robotic manipulation. A cerebellum-inspired pre-trained neural network has been implemented as a forward model as a means of circumventing potential problems with traditional feedback controllers for space manipulators. Such problems are anticipated in sophisticated real-world applications of space manipulation such as the manipulator-mounted free-flyer spacecraft. Cerebellar-inspired forward models have thus far been demonstrated in only simple proof-of-concept problems. A novel method was developed for combining the multi-layer perceptron (MLP) with a multi-output regression tree (MORT) in a pre-trained feedforward neural network capable of predicting forward trajectories to an accuracy of 89-96% for previously unseen trajectory data. Given the similarity in form and dynamics between earth-based and space-based robotic manipulators, transfer learning of the neural network controller was explored as an avenue to address the challenges of experimental simulation of the space environment on earth and addressing the limited computation resources onboard spacecraft. This research discovered a practical limitation of such transfer learning and neural networks in general as a model of general intelligence. Consequently, alternative approaches were investigated. A reinforcement learning approach has been selected based on learning an adaptive model. This model can be applied to the space environment by leveraging rapid motor adaptation. A pre-trained feedforward model using online updates from proprioceptive data can be implemented on limited computational resources. Unlike other pre-trained forward models, it does not require online re-computation for new environments. The pre-trained forward model incorporates compliance in a force control architecture. A future task would be the experimentation of different types of payloads using some muscle-like compliant approach to make proprioceptive measurements.
Degree
thesis:*- Name thesis:degree_name
- Doctor of Philosophy (Ph.D.)
- Level thesis:degree_level
- Doctoral
- Discipline thesis:degree_discipline
- Engineering, Aerospace
- Grantor dc:publisher
- Carleton University
- Year dc:date.issued
- 2024
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Ogundipe, Collins Olusegun Abiodun
Subjects
dc:subject × 1Rights
dc:rights- Statement dc:rights
-
- Copyright © 2024 the author(s). Theses may be used for non-commercial research, educational, or related academic purposes only. Such uses include personal study, distribution to students, research and scholarship. Theses may only be shared by linking to the Carleton University Institutional Repository and no part may be copied without proper attribution to the author; no part may be used for commercial purposes directly or indirectly via a for-profit platform; no adaptation or derivative works are permitted without consent from the copyright owner.
- Language dc:language.iso
- en