Robust Coordination of Autonomous Systems through Risk-sensitive, Model-based Programming and Execution
Abstract
Unlike their human counterparts, most autonomous systems to date are not effective at characterizing or bounding mission risk. In this project, we enabled the development of risk-sensitive autonomous systems through three main contributions: first, we introduced cRMPL, an extension of RMPL where one can specify acceptable risk levels for different mission segments through the addition of chance constraints. Second, we extended the continuous planner, used by our executive, to generate and adapt plans that maximize expected utility within the risk bounds specified by the operators. Planning is performed through novel stochastic optimization algorithms that allocate user-specified risk to actions and constraints according to the benefit received. We evaluated the generality of this risk-sensitive paradigm in simulation and hardware, for autonomous air or space vehicles and humanoid logistics support robots. Benefits include increased number and complexity of vehicle missions for a fixed operational cost, increased robot safety around humans; a reduction in unacceptable mission failure or robot loss, and improved mission return within defined risk levels.
Document Details
- Document Type
- Technical Report
- Publication Date
- Oct 09, 2015
- Accession Number
- ADA627055
Entities
People
- Brian Williams
- Cheng Fang
- Eric Timmons
- Pedro Santana
Organizations
- Massachusetts Institute of Technology