Scalable Multi-Robot System for Non-myopic Spatial Sampling

Abstract

This paper presents a distributed scalable multirobot planning algorithm for non-uniform sampling of quasistatic spatial fields. We address the problem of efficient data collection using multiple autonomous vehicles. In this paper, we are interested in analysing the effect of communication between multiple robots, acting independently, on the overall sampling performance of the team. Our focus is on distributed sampling problem where the robots are operating independent of their teammates, but have the ability to communicate their states to other neighbors with a constraint on the communication range. We design and apply an informed non-myopic path planning technique on multiple robotic platforms to efficiently collect measurements from a spatial field. Our proposed approach is highly adaptive to challenging environments, growing team size, and runs in real-time, which are the key features for any realworld scenario. The results show that our distributed sampling approach is able to achieve efficient sampling with minimal communication between the robots. We evaluate our approach in simulation over multiple distributions commonly occurring in nature and on the real-world data collected during a field trial.

MARLAS: Multi Agent Reinforcement Learning for cooperated Adaptive Sampling

“…In Fig. 3, we compare MARLAS with multiple baseline sampling methods including independently trained multi-robot sampler [19], DARP (Divide Area based on the Robot's initial Positions) algorithm [15], and maxima search algorithm [22]. We used discounted accumulated reward and average pair-wise overlap metrics for these comparisons.…”

“…In our previous work [3,19], we proposed and evaluated various feature aggregation methods for adaptive sampling. Multiresolution feature aggregation [3] has been empirically proven to outperform other methods, as it reduces the state dimension and utilizes the geometry induced bias for the sampling task.…”