Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Data-driven Spatial Classification using Multi-Arm Bandits for Monitoring with Energy-Constrained Mobile Robots

Xiaoshan Lin, Siddharth Nayak, Stefano Di Cairano, Abraham P. Vinod

TL;DR

This work addresses rapid spatial classification for monitoring with a heterogeneous, energy-constrained team of mobile sensors and charging stations. It introduces a bi-level framework that uses a bandit-based high-level planner to select epoch goals from the candidate set $\mathscr{C}$ and an IP-based low-level planner to coordinate collision-free trajectories under energy constraints, with collision avoidance refined by a linear assignment step. The approach provides anytime guarantees and finite-time bounds, and its effectiveness is demonstrated through hardware experiments with drones and ground robots as well as extensive simulations in noisy settings. The results indicate strong practical viability for applications such as search-and-rescue and environmental monitoring, offering scalable, data-driven planning under uncertainty.

Abstract

We consider the spatial classification problem for monitoring using data collected by a coordinated team of mobile robots. Such classification problems arise in several applications including search-and-rescue and precision agriculture. Specifically, we want to classify the regions of a search environment into interesting and uninteresting as quickly as possible using a team of mobile sensors and mobile charging stations. We develop a data-driven strategy that accommodates the noise in sensed data and the limited energy capacity of the sensors, and generates collision-free motion plans for the team. We propose a bi-level approach, where a high-level planner leverages a multi-armed bandit framework to determine the potential regions of interest for the drones to visit next based on the data collected online. Then, a low-level path planner based on integer programming coordinates the paths for the team to visit the target regions subject to the physical constraints. We characterize several theoretical properties of the proposed approach, including anytime guarantees and task completion time. We show the efficacy of our approach in simulation, and further validate these observations in physical experiments using mobile robots.

Data-driven Spatial Classification using Multi-Arm Bandits for Monitoring with Energy-Constrained Mobile Robots

TL;DR

This work addresses rapid spatial classification for monitoring with a heterogeneous, energy-constrained team of mobile sensors and charging stations. It introduces a bi-level framework that uses a bandit-based high-level planner to select epoch goals from the candidate set and an IP-based low-level planner to coordinate collision-free trajectories under energy constraints, with collision avoidance refined by a linear assignment step. The approach provides anytime guarantees and finite-time bounds, and its effectiveness is demonstrated through hardware experiments with drones and ground robots as well as extensive simulations in noisy settings. The results indicate strong practical viability for applications such as search-and-rescue and environmental monitoring, offering scalable, data-driven planning under uncertainty.

Abstract

We consider the spatial classification problem for monitoring using data collected by a coordinated team of mobile robots. Such classification problems arise in several applications including search-and-rescue and precision agriculture. Specifically, we want to classify the regions of a search environment into interesting and uninteresting as quickly as possible using a team of mobile sensors and mobile charging stations. We develop a data-driven strategy that accommodates the noise in sensed data and the limited energy capacity of the sensors, and generates collision-free motion plans for the team. We propose a bi-level approach, where a high-level planner leverages a multi-armed bandit framework to determine the potential regions of interest for the drones to visit next based on the data collected online. Then, a low-level path planner based on integer programming coordinates the paths for the team to visit the target regions subject to the physical constraints. We characterize several theoretical properties of the proposed approach, including anytime guarantees and task completion time. We show the efficacy of our approach in simulation, and further validate these observations in physical experiments using mobile robots.
Paper Structure (16 sections, 3 theorems, 14 equations, 6 figures, 4 tables, 2 algorithms)

This paper contains 16 sections, 3 theorems, 14 equations, 6 figures, 4 tables, 2 algorithms.

Table of Contents

  1. Introduction
  2. Problem statement
  3. Methodology
  4. High-level Planner: Bandit-based decision making
  5. Low-level coordination: Integer Program-based solution
  6. Collision avoidance for the sensors
  7. Theoretical guarantees
  8. Hardware-based validation
  9. Simulation-based performance analysis
  10. Sensor accuracy (varying the support of $\mu_l$)
  11. Number of epoch goals $D$
  12. Agility of the team
  13. Scalability
  14. Conclusion
  15. Proof of Proposition \ref{['prop:anytime']}
  16. ...and 1 more sections

Key Result

Proposition 1

At any epoch $p\in\mathbb{N}$, the sets $\mathcal{K}(p)$ and $\mathcal{R}(p)$ in Algorithm algo:high_level satisfy $\mathcal{K}(p)\subseteq\mathcal{S}_{\theta-\epsilon}$ and $\mathcal{R}(p)\subseteq\mathscr{C}\setminus\mathcal{S}_{\theta+\epsilon}$, with probability of at least $1-\delta$.

Figures (6)

  • Figure 1: Data-driven multi-agent search under noisy observations. A high-level planner uses the available measurements of cells to determine the confidences of an unclassified cell to be interesting, and determines the cells to visit next. A low-level planner computes trajectories for the agents subject to various motion constraints arising from dynamics as well as energy limitations. Altogether, the proposed approach guarantees safe, data-driven sensor deployment to address Problem \ref{['prob_st:main']}.
  • Figure 2: (Top) Four types of potential collisions in the paths computed by the low-level motion planner. (Bottom) Alternative collision-free paths that visit the same waypoints as in the top figure, identified via a linear assignment.
  • Figure 3: Snapshot of the hardware experiment with two Turtlebot4 robots as stations and four Crazyflie drones as sensors. The colored triangular prisms represent the obstacles or no-fly zones for the sensors. The second column and the third row (delimited by small triangular tiles) represent the roads where the stations must stay. The four green tiles represent the interesting cells, i.e., they contain the search objective. All other cells are considered uninteresting. See video of the physical experiment at https://youtu.be/gzulpOcVYzg.
  • Figure 4: Classification progress of Algorithm \ref{['algo:high_level']} in the hardware experiments with the unmodified and degraded sensors.
  • Figure 5: Classification progress of Algorithm \ref{['algo:high_level']} when varying worst-case sensor accuracy $\mu_{l,\text{interest}}^\text{worst}$. Epoch (x-axis) is shown in log-scale. The epochs needed to classify all cells (vertical dashed lines) decreases with increasing sensor accuracy.
  • ...and 1 more figures

Theorems & Definitions (3)

  • Proposition 1: Anytime algorithm
  • Proposition 2: Finite time guarantees
  • Corollary 1