Summary
Autonomous robots such as autonomous vehicles cars and drones have the potential to revolutionize the way we work and live. Unfortunately, current autonomous robots do not have ‘common sense’ and may enter a catastrophic condition called loss-of-control after failures. The ultimate goal of this research is to enable a new generation of autonomous robots that are aware of their physical capabilities and limitations, allowing them to act with common sense after failures.
To achieve this I propose a new paradigm in autonomous robot control: ‘Artificial Physical Awareness’ (APA). APA requires accurate real-time knowledge of the time-varying stochastic safe envelope which is a subset of the state-space inside which safe operations of the autonomous robot can be guaranteed. The safe envelope is stochastic and time-varying; it contains uncertainties and will shrink after failures, reflecting the reduced post-failure performance of the autonomous robot.
Obtaining and utilizing the time-varying stochastic safe envelope in real-time represents a currently unsolved scientific challenge for the following reasons: 1) the safe envelope cannot be measured directly; 2) current safe envelope computation methods are real-time intractable and/or do not take into account uncertainties; and 3) no control methodology exists that allows for time-varying safe-envelope informed balancing of safety and performance.
This multidisciplinary research combines new insights in time-varying stochastic state reachability analysis, tipping-point forecasting, bio-inspired envelope sensing and recovery, and nonlinear fault-tolerant control to develop the new APA-autopilot system, which is the main output of this research. This research project has the potential to lead to a revolution in autonomous robot design, and operations by providing transparent safety and performance bounds, even after failures.
To achieve this I propose a new paradigm in autonomous robot control: ‘Artificial Physical Awareness’ (APA). APA requires accurate real-time knowledge of the time-varying stochastic safe envelope which is a subset of the state-space inside which safe operations of the autonomous robot can be guaranteed. The safe envelope is stochastic and time-varying; it contains uncertainties and will shrink after failures, reflecting the reduced post-failure performance of the autonomous robot.
Obtaining and utilizing the time-varying stochastic safe envelope in real-time represents a currently unsolved scientific challenge for the following reasons: 1) the safe envelope cannot be measured directly; 2) current safe envelope computation methods are real-time intractable and/or do not take into account uncertainties; and 3) no control methodology exists that allows for time-varying safe-envelope informed balancing of safety and performance.
This multidisciplinary research combines new insights in time-varying stochastic state reachability analysis, tipping-point forecasting, bio-inspired envelope sensing and recovery, and nonlinear fault-tolerant control to develop the new APA-autopilot system, which is the main output of this research. This research project has the potential to lead to a revolution in autonomous robot design, and operations by providing transparent safety and performance bounds, even after failures.
Unfold all
/
Fold all
More information & hyperlinks
| Web resources: | https://cordis.europa.eu/project/id/101126132 |
| Start date: | 01-07-2024 |
| End date: | 30-06-2029 |
| Total budget - Public funding: | 1 996 040,00 Euro - 1 996 040,00 Euro |
Cordis data
Original description
Autonomous robots such as autonomous vehicles cars and drones have the potential to revolutionize the way we work and live. Unfortunately, current autonomous robots do not have ‘common sense’ and may enter a catastrophic condition called loss-of-control after failures. The ultimate goal of this research is to enable a new generation of autonomous robots that are aware of their physical capabilities and limitations, allowing them to act with common sense after failures.To achieve this I propose a new paradigm in autonomous robot control: ‘Artificial Physical Awareness’ (APA). APA requires accurate real-time knowledge of the time-varying stochastic safe envelope which is a subset of the state-space inside which safe operations of the autonomous robot can be guaranteed. The safe envelope is stochastic and time-varying; it contains uncertainties and will shrink after failures, reflecting the reduced post-failure performance of the autonomous robot.
Obtaining and utilizing the time-varying stochastic safe envelope in real-time represents a currently unsolved scientific challenge for the following reasons: 1) the safe envelope cannot be measured directly; 2) current safe envelope computation methods are real-time intractable and/or do not take into account uncertainties; and 3) no control methodology exists that allows for time-varying safe-envelope informed balancing of safety and performance.
This multidisciplinary research combines new insights in time-varying stochastic state reachability analysis, tipping-point forecasting, bio-inspired envelope sensing and recovery, and nonlinear fault-tolerant control to develop the new APA-autopilot system, which is the main output of this research. This research project has the potential to lead to a revolution in autonomous robot design, and operations by providing transparent safety and performance bounds, even after failures.
Status
SIGNEDCall topic
ERC-2023-COGUpdate Date
02-10-2024
Images
No images available.
Geographical location(s)
