Multi-modal active perception for information gathering in science missions

Abstract

Robotic science missions in remote environments, such as deep ocean and outer space, can involve studying phenomena that cannot directly be observed using on-board sensors but must be deduced by combining measurements of correlated variables with domain knowledge. Traditionally, in such missions, robots passively gather data along prescribed paths, while inference, path planning, and other high level decision making is largely performed by a supervisory science team. However, communication constraints hinder these processes, and hence the rate of scientific progress. This paper presents an active perception approach that aims to reduce robots' reliance on human supervision and improve science productivity by encoding scientists' domain knowledge and decision making process on-board. We use Bayesian networks to compactly model critical aspects of scientific knowledge while remaining robust to observation and modeling uncertainty. We then formulate path planning and sensor scheduling as an information gain maximization problem, and propose a sampling-based solution based on Monte Carlo tree search to plan informative sensing actions which exploit the knowledge encoded in the network. The computational complexity of our framework does not grow with the number of observations taken and allows long horizon planning in an anytime manner, making it highly applicable to field robotics. Simulation results show statistically significant performance improvements over baseline methods, and we validate the practicality of our approach through both hardware experiments and simulated experiments with field data gathered during the NASA Mojave Volatiles Prospector science expedition.

Adaptive sampling of transient environmental phenomena with autonomous mobile platforms

“…In order to compactly represent and plan with continuous states and observations, it is common in literature to discretize the state space [14,15]. In the MSS problem, discretization over space and time is complicated by the reward function, which requires precision about the location of the global maximum.…”

Untitled

“…Example 2: An astrogeologist is studying the geology in a Martian crater, as in Arora et al [3]. A discrete grid may be used to represent the environment of interest,where hyper-spectral camera observations can be taken in each grid cell to infer the geologic type of that cell.…”

“…Bayesian networks have seen adoption in scientific sensing problems, as they allow for a very flexible environmental model and directly encode causal structure between the random variables in the environment using the dependency graph. Arora et al [3] model the environment for scientific information gathering as a discrete grid and use a tree-structured BN to represent the geologic type of a cell based on nearby cells and environmental sensor measurements. Candela et al [14] use a BN for hypothesisdriven geologic mapping using a mobile robot in a space-based exploration application.…”

“…Zheng et al [123] derive a sampling-based method for estimating mutual information in general models with performance guarantees. Arora et al [3] apply a Monte Carlo sampling approach to estimate mutual information without performance guarantees in general Bayesian network environmental models. The utility of information-theoretic reward measures is evidenced by their widespread adoption in these diverse application areas.…”

“…Although this chapter does not explore how this model can be used within a scientific planning framework, we propose the use of model perplexity as a measure of the information content of a scene and evaluate how well measures of perplexity map to human intuition about scientifically interesting scenes. The modeling problem is introduced within the context of autonomous deep sea exploration in Section 4.1, followed by a review of related literature in Section 4.2.The hybrid approach employing Bayesian topic models and deep convolutional autoencoders is presented in Section 4 3…”