Skip to main content

Communicating agent intentions for human-agent decision making under uncertainty.

Porteous, J., Lindsay, A. and Charles, F., 2023. Communicating agent intentions for human-agent decision making under uncertainty. In: The 22nd International Conference on Autonomous Agents and Multiagent Systems, 29 May to 2 June 2023, London, UK. (In Press)

Full text available as:

[img] PDF
AAMAS2023_JPorteous_SubmittedVersion.pdf - Accepted Version
Restricted to Repository staff only until 30 May 2023.
Available under License Creative Commons Attribution Non-commercial No Derivatives.

3MB

Official URL: https://aamas2023.soton.ac.uk/

Abstract

Recent advances in visualisation technologies have opened up new possibilities for human-agent communication. For systems where agents use automated planning, visualisation of agent intentions, i.e., agent planned actions, can assist human understanding and decision making (e.g., deciding when human control is required or when it can be delegated to an agent). We are working in an application area, shipbuilding, where branched plans are often essential, due to the typical uncertainty experienced. Our focus is how best to communicate, using visualisation, the key information content of branched plans. It is important that such visualisations communicate the complexity and variety of the possible agent intentions i.e., executions, captured in a branched plan, whilst also connecting to the practitioner’s understanding of the problem. Thus we utilise an approach to generate the complete branched plan, to be able to provide a full picture of its complexity, and a mechanism to select a subset of diverse traces that characterise the possible agent intentions. We have developed an interface which uses 3D visualisation to communicate details of these characterising execution traces. Using this interface, we conducted a study evaluating the impact of different modes of presentation on user understanding. Our results support our expectation that visualisation of branched plan characterising execution traces increases user understanding of agent intention and plan execution possibilities.

Item Type:Conference or Workshop Item (Paper)
Uncontrolled Keywords:Virtual Agents; Human-Agent Decision Making; Explainable AI Planning
Group:Faculty of Science & Technology
ID Code:38053
Deposited By: Symplectic RT2
Deposited On:10 Feb 2023 15:52
Last Modified:10 Feb 2023 15:57

Downloads

Downloads per month over past year

More statistics for this item...
Repository Staff Only -