Expert-Informed Autonomous Science Planning for In-situ Observations and Discoveries

McMahon, Jay W.; Ahmed, Nisar; Lahijanian, Morteza; Peter, Amorese,; Deka, Taralicin; Muvvala, Karan; Slack, Trevor; Wakayama, Shohei

doi:10.1109/aero53065.2022.9843445

Cited by 4 publications

(3 citation statements)

References 35 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…the sum of difference between the maximum expected outcome and an actual outcome over iterations, via alternately performing two steps: (1) action selection: selecting one option (a.k.a "arm") from alternatives with unknown outcome distributions; and (2) observation update: updating the statistics for the selected option based on the outcome obtained. In stochastic bandits, it is common to apply the Bayes' rule to update the posterior of the estimated outcome distribution of the selected option in step (2). For step (1), various action selection methods such as -greedy, the upper confidence bound (UCB) [15], and the Thompson sampling (TS) [17] have been devised.…”

Section: Background and Related Work A Multi Armed Bandits (Mabs) And...mentioning

confidence: 99%

“…NASA recently launched the Concepts for Ocean worlds Life Detection Technology (COLDTech) program to research and develop software for 'fail-operational' autonomous robotic landers that will be dispatched to icy solar system moons like Europa and Enceladus [2]. Currently, the concept of operations calls for a robotic lander to perform basic surface tasks such as digging, imaging, moving objects, retrieving/analyzing samples with various instruments, and downlinked selected science data.…”

Section: A Motivating Problem Scenariomentioning

confidence: 99%

“…mission time and communication constraints (e.g. due to high radiation and hours long round trip time delays for speed of light signals) make smooth human-robot interaction hard in such missions [1], [2]. Thus, the robot must autonomously and accurately infer the outcome probability that each area contains valuable objects, while using sensors that are easily deployable and consume little energy.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Active Inference for Autonomous Decision-Making with Contextual Multi-Armed Bandits

Wakayama¹,

Ahmed²

2022

Preprint

Self Cite

View full text Add to dashboard Cite

In autonomous robotic decision-making under uncertainty, the tradeoff between exploitation and exploration of available options must be considered. If secondary information associated with options can be utilized, such decision-making problems can often be formulated as a contextual multi-armed bandits (CMABs). In this study, we apply active inference, which has been actively studied in the field of neuroscience in recent years, as an alternative action selection strategy for CMABs. Unlike conventional action selection strategies, it is possible to rigorously evaluate the uncertainty of each option when calculating the expected free energy (EFE) associated with the decision agent's probabilistic model, as derived from the free-energy principle. We specifically address the case where a categorical observation likelihood function is used, such that EFE values are analytically intractable. We introduce new approximation methods for computing the EFE based on variational and Laplace approximations. Extensive simulation study results demonstrate that, compared to other strategies, active inference generally requires far fewer iterations to identify optimal options and generally achieves superior cumulative regret, for relatively low extra computational cost.

show abstract