Collaborative Macroscopic Ensemble Modeling and Control of Robot Teams
We present solutions to the multi robot task allocation (MRTA) problem by controlling robot-robot collaborations to inform time-varying team-wide objectives at the macroscopic level.
Research Objective:
Environmental monitoring requires assigning robots to sampling locations. This is a variant of the Multi Robot Task Allocation (MRTA) problem. We know that some dynamic environments have known periodic behavior, e.g., rivers with tidal shifts. Robots must perform coordinated sampling strategies at specific moments in time which is categorized as the scheduling and coalition formation problem. Existing microscopic solutions plan and control for individual robots and require expensive replanning, e.g., when environment or task conditions change. Interestingly, biologists have long studied how animal populations make fast resource selections in dynamic and uncertain environments. One popular albeit simple model of this phenomena is referred to as the ideal free distribution (IFD) model: animals make random weighted resource selection based on the perceived resource value. The key is that each individual may select an option that is not optimal for the individual but instead beneficial for the overall survival of the population. Taking inspiration from biology, macroscopic ensemble methods model populations of robots where each robot makes a weighted random task selection based on model parameters. Macroscopic approaches have been shown to have nice analytical guarantees, scalability, and can even be applied to heterogeneous robot collectives. However, macroscopic methods often ignore robot-robot collaborations in favor of model simplicity, despite this being a key feature of existing biological models. Existing macroscopic models are asymptotically stable, i.e., given a desired distribution of robots we can find model parameters that will ensure those populations are achieved. The problem we aim to address is: design a macroscopic ensemble modeling and control method for multi robot teams that controls time-varying populations of robots. Our solution introduces robot-robot collaborations to the model which allow for the potential of time-varying distributions of robots in the workspace. Furthermore, our latest results show a model reinterpretation that allows the break-down of the well-mixed assumption. This means we can achieve time-varying populations without needing robots to perform tasks throughout the environment. Our latest experimental results are shown below!
We experimentally verify our methods using the miniature Autonomous Surface Vehicles (mASVs) which are built in house at the ScalAR lab. Our testing environment is a 4m x 3m x 1.5m tank, equipped with 13 OptiTrack cameras, and the capability to make gyres in the tank. At any given time, we can have upwards of 10 mASV in the tank performing different control strategies. Within the tank we have demonstrated heterogeneous teams using Crazyflie robots and mASV simultaneously.
Relevant Publications:
Recently our work was accepted to Autonomous Robotics! Building off of prior results for this project, we take inspiration from population modeling in other disciplines and present a nonlinear macroscopic ensemble allocation model that describes individual robot collaboration with the potential for time-varying task assignment without the need for replanning populations. Our results demonstrate a range of possible time-varying task assignment behaviors that are potential solutions to handling known periodic environments or task changes. In addition, we explore the breakdown of classic macroscopic modeling assumptions and present model reinterpretations to mitigate their impact. Our simulation and experimental results demonstrate time-varying task assignment, which could be applied to tasks like environmental monitoring, collective construction, and resource distribution.
I recently presented at DARS ‘24 at Cornell Tech on collaborative macroscopic ensemble modeling and control of robot teams. We showed that our proposed model had the possibility of capturing time-varying populations and we introduced two interpretations of robot-robot collaboration. This work had an emphasis on studying the well-mixed assumption which is critical when using macroscopic ensemble methods. Our work was well received by the audience and I had a great round of questions after my talk!
I presented our work to IROS ‘23 in Detroit, MI. This work covered the application of collaborative macroscopic ensemble methods to environmental monitoring scenarios. Our solution avoids partitioning the environment and allows the robots to more flexibly perform the monitoring task.
Our work presented at DARS ‘22 uses a nonlinear stochastic model to achieve time-varying distributions of robots which was a limitation of previous linear stochastic models. Our results suggest further work is needed to understand how macroscopic models can better incorporate feedback from the environment.
This work is done in collaboration with Dr. Thales C. Silva and Dr. M. Ani Hsieh at the University of Pennsylvania GRASP lab.