Future space exploration missions will require robotic assistants that adapt to complex, rapidly changing environments. NASA's Autonomous Systems and Operations (ASO) project is applying intelligent agents that plan future activities using a predictive model of the environment. However, this model rarely matches reality exactly, and this makes their plans difficult to execute in the real world. Two major issues in the execution of AI plans are: how to handle temporal coordination in the face of uncertain processes, and how to handle non-determinism (the chance that an action might fail, or produce unexpected results).
The goal of this project is to develop algorithms for the execution of plans. This involves: investigating existing approaches (such as temporal plan networks); implementing an algorithm that will decide when to execute each action of a plan; and evaluating the algorithm in different simulations.
The output of this project has the potential for concrete application in real scenarios, such as: (LINK)Situational Awareness Models for Human-AI Teaming
Intelligent Agents often plan their future behaviour using a predictive model of the environment. When interacting with humans, the future becomes very hard to predict. One way to cope with this uncertainty is through contingencies: choice-points in plans that allow for branching behaviour based on the outcome of sensing or human input.
The goal of this project is to generate a predictive contingent model that can be used by a mixed team of AI and human agents operating in collaboration with a human supervisor. The supervisor is remote and only able to monitor the information that is presented by the team. The team will have to find a balance between completing their tasks efficiently, and giving the supervisor enough information to understand the situation. At the same time, the supervisor might interrupt the team with new information, such as additional objectives.
As an additional goal for a more ambitious project, the model can be embedded into an existing system for disaster recovery and search & rescue.A Simulation of Multi-Agent Collaboration
Multi-agent scenarios with limited communication present complex and challenging problems for AI. While single agents are able to plan complex behaviour for themselves, agents operating in teams often fall back on pre-scripted behaviours to more easily facilitate collaborative action.
The goal of this project is threefold: (a) to investigate how incoming tasks are dynamically assigned to teams of AI agents, (b) to investigate the relationship between the restriction on inter-agent communication and the limit on collaborative action, and (c) to develop a simulation in which multiple autonomous robots must collaborate to achieve a goal.
A successful project could make use of existing materials to be used as the foundation of the proposed Planning and Execution Competition for Logistics Robots In Simulation at RoboCup 2020 (based on the Logistics League LINK).