Georgia Institute of Technology
Coordination of Robot Swarms in Crowded and Confined Conditions
Abstract
dc:description.abstractThe intricate task of coordinating robot swarms in highly constrained and crowded environment is often hindered by the robots’ limited capability to gather precise state information of their environment crucial for controlling the group toward their objectives. Such challenges stem from inherent uncertainties and unpredictability within these environments, influencing the interactions between agents and their surroundings. However, studies of dense active matter reveal that coordinated behaviors could arise through self-organization principles, where complex global configurations and patterns emerge from simple, local, and social interactions. This thesis aims to exploit these unavoidable features of dense active matter to enable the spontaneous coordination of robot swarms in the absence of global control, and explicit information about the state of other robots. Specifically, an adaptive clog control technique facilitates the emergent coordination of robot swarms through unplanned contact interactions. This technique relies on inter-robot contacts and collisions amid noisy estimations of their locations and environmental changes. Persistent clogs in narrow tunnels become resolved through voluntary directional changes and individual participation modulation within group tasks. Next, we improved and extended our algorithm to handle scenarios where an individual robot’s failure could detrimentally impact group performance. The Active Contact Response (ACR) algorithm imparts fault tolerance within the clog control mechanism, allowing the repositioning of faulty robots to less obstructive configurations, ensuring continuous traffic flow within the tunnel and completion of tasks. Additionally, these characteristics of crowded and confined systems address foundational challenges in multi-agent reinforcement learning. By establishing an indirect communication channel through the shared environment, agents can estimate and reason about the statistics of other agents’ activities, directly reducing the environment’s non-stationarity from an individual agent’s perspective. Numerical simulations show the potential benefits of leveraging local and physical interactions, and implicit communication via the environment to enhance convergence and scalability within multi-agent reinforcement learning setting. Finally, a mathematical model was formulated to elucidate and validate the self-organizing behavior observed in naturally occurring swarms navigating crowded and narrow self-generated tunnels without relying on global control or explicit communication. The model’s alignment with multi-phase excavation rates observed in biological experiments suggests the pivotal role of contact and local interactions in governing the self-regulation behaviors exhibited by natural collectives.
Degree
thesis:*- Level thesis:degree_level
- Doctoral
- Department dc:contributor.department
- Electrical and Computer Engineering
- Grantor dc:publisher
- Georgia Institute of Technology
- Year dc:date.issued
- 2024
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Aina, Kehinde
- Advisor dc:contributor.advisor
-
- Goldman, Daniel I.
- Committee members dc:contributor.committeemember
-
- Ha, Sehoon
- Hutchinson, Seth
- Randall, Dana
- Hu, David
Subjects
dc:subject × 11Rights
- Language dc:language.iso
- en_US
Identifiers
dc:identifier.*- Handle dc:identifier.uri
- https://hdl.handle.net/1853/75337