Human-Collective Collaborative Site Selection

TL;DR

This paper evaluates two collective decision models for human-robot teams in complex tasks, demonstrating that a bias-reducing model improves accuracy and reduces operator influence, with a novel interaction strategy for adjusting autonomy.

Contribution

It introduces an extended bias-reducing decision model and a new human-collective interaction strategy, enhancing accuracy and operator efficiency in collective decision-making.

Findings

Extended model is 57% more accurate in difficult decisions.

Human-collective teams with the bias-reducing model achieve 25% higher accuracy.

The novel interaction strategy allows dynamic adjustment of collective autonomy.

Abstract

Robotic collectives are large groups (at least 50) of locally sensing and communicating robots that encompass characteristics of swarms and colonies, whose emergent behaviors accomplish complex tasks. Future human-collective teams will extend the ability of operators to monitor, respond, and make decisions in disaster response, search and rescue, and environmental monitoring problems. This manuscript evaluates two collective best-of-n decision models for enabling collectives to identify and choose the highest valued target from a finite set of n targets. Two challenges impede the future use of human-collective shared decisions: 1) environmental bias reduces collective decision accuracy when poorer targets are easier to evaluate than higher quality targets, and 2) little is understood about shared human-collective decision making interaction strategies. The two evaluated collective best-of-n models include an existing insect colony decision model and an extended bias-reducing model that attempts to reduce environmental bias in order to improve accuracy. Collectives using these two strategies are compared independently and as members of human-collective teams. Independently, the extended model is slower than the original model, but the extended algorithm is 57% more accurate in decisions where the optimal option is more difficult to evaluate. Human-collective teams using the bias-reducing model require less operator influence and achieve 25% higher accuracy with difficult decisions, than the human-collective teams using the original model. Further, a novel human-collective interaction strategy enables operators to adjust collective autonomy while making multiple simultaneous decisions.