MIT Engineers Develop Algorithm to Identify and Fix Failures in Autonomous Systems
MIT engineers have created an approach that can be used with any autonomous system to quickly detect a variety of potential failures before they are deployed in the real world. This algorithm can also find fixes to these failures and suggest repairs to prevent system breakdowns. The team has proven that this approach, in the form of an automated sampling algorithm, is effective in identifying possible failures in simulated autonomous systems, including power grids, aircraft collision avoidance systems, rescue drones, and robotic manipulators.
Sensitivity over Adversaries
Prompted by the Texas power crisis in 2021, the team focused on expanding their research to identify and fix failures in more complex, large-scale autonomous systems. The team developed an algorithm that automatically generates random changes within a system, assesses the system’s sensitivity, and identifies a wider range of potential failures. This sensitive approach enables the researchers to discover possible failures and identify fixes.
Hidden Failures
The team has tested the new approach in simulated autonomous systems, and it has proven to be effective. For example, in the case of a power grid, the algorithm identified the potential for a complete blackout when combined with the failure of a second power line, something that conventional approaches had missed. The researchers also demonstrated the effectiveness of this approach on a robotic arm that successfully executed a task after implementing the suggested fix from the algorithm.
The team’s approach could be used with any autonomous system as long as it comes with an accurate simulation of its behavior. The goal is to develop this approach into an app that designers and engineers can use to fine-tune and improve their systems before testing in the real world.
The research is supported by NASA, the National Science Foundation, and the U.S. Air Force Office of Scientific Research.