Energy-Aware Coverage Planning for Heterogeneous Multi-Robot System

TL;DR

A new energy-aware controller based on Lloyd's algorithm to adapt the weights of the robots based on their energy dynamics and divide the area of interest among the robots accordingly is proposed.

Abstract

We propose a distributed control law for a heterogeneous multi-robot coverage problem, where the robots could have different energy characteristics, such as capacity and depletion rates, due to their varying sizes, speeds, capabilities, and payloads. Existing energy-aware coverage control laws consider capacity differences but assume the battery depletion rate to be the same for all robots. In realistic scenarios, however, some robots can consume energy much faster than other robots; for instance, UAVs hover at different altitudes, and these changes could be dynamically updated based on their assigned tasks. Robots' energy capacities and depletion rates need to be considered to maximize the performance of a multi-robot system. To this end, we propose a new energy-aware controller based on Lloyd's algorithm to adapt the weights of the robots based on their energy dynamics and divide the area of interest among the robots accordingly. The controller is theoretically analyzed and extensively evaluated through simulations and real-world demonstrations in multiple realistic scenarios and compared with three baseline control laws to validate its performance and efficacy.

Figures (9)

• Figure 1: Regions assigned by standard Voronoi partitioning (left) and the proposed energy-aware controller (EAC) (right). The robots have the same initial battery level. However, robot 3's depletes its energy three times faster than the other robots. Therefore, robot 3's EAC-assigned region area (area in $m^2$ in parentheses) is less than other robots by adapting the weights ($w_i$) based on the ratio of energy depletion rate between the robots.
• Figure 2: Results of Scenario 1, where all robots have the same energy characteristics $E_i(0)=100, \alpha_i=1, \beta_i=1$, but robot $5$ has a higher temporal energy depletion rate $\alpha_5=5$. The top plots show the initial configurations and the final partitions calculated by all the approaches. The bottom plots show the cost comparison (left-most) and the area convergence over time.
• Figure 3: Time-evolution of the robot energy levels $E_i$, weights $w_i$, and the convergences of weight ratios $w_i\dot{E}_i$ and the robot distance from the centroid of their respective Voronoi cell in Scenario 1 with EAC Controller.
• Figure 4: Results of Scenario 2 experiments, where robots have a heterogeneous combination of energy levels and depletion rates $\alpha_4=4$ and $E_4^{init}=100$, where other robots have $\alpha = 1, E^{init}=25$. Because robot $4$ started with high capacity but depleted energy at a 4.4x rate compared to other robots, their dynamics would eventually cancel out each other, and the final weights should remain similar for all robots in an energy-aware coverage.
• Figure 5: Results of Scenario 3 experiments, showing the effect of dynamic $\beta(t)$. Initially, robots 3 and 4 had $\beta=5$, while other robots have $\beta=1$. This rate flipped at time $t=11$, and again reset to initial values at $t=22$. EAC and ATC adapted to these dynamics in energy depletion rates.

Theorems & Definitions (6)

• theorem 1
• proof : Theorem 1
• corollary 1
• proof
• theorem 2
• proof