Unmet Need
Currently, 55% of the world’s population lives in urban areas, a proportion that is expected to grow to 68% by 2050 (United Nations, 2018). The increasing growth of urban population will put heavy stress on existing infrastructure such as the transportation system. Outdated traffic signal timing accounts for more than 10% of all traffic delay and congestion on major routes alone (US Department of Transportation, 2017). Implementing “smart traffic controllers” using artificial intelligence has the potential to alleviate the existing and future issues regarding traffic in major cities. However, artificial intelligence is vulnerable to edge failure cases that can have both performance and safety ramifications (Computer Vision and Pattern Recognition, 2014). Therefore, there is a strong need for artificial intelligence systems that can be reliably monitored and adjusted to ensure safe and effective control.
Technology Overview
Johns Hopkins researchers have developed a system for monitoring artificial intelligence (AI) controller performance that can change control policies if the controller is in danger of failure. By monitoring the confidence of the AI controller, the Runtime Assurance of Distributed Intelligent Control Systems (RADICS) approach is capable of quickly switching away from the AI controller to a safer control policy. Once the potential crisis has been averted, RADICS gives control back to the AI controller for an overall improvement in performance without sacrificing safety during failure cases.
Stage of Development
The invention has been evaluated in a popular Traffic Light controller framework (Simulation of Urban Mobility) and proven to be more effective at handling failure cases than current AI solutions.
Publication
B. Wheatman, J. Chen, T. Sookoor and Y. Amir, "RADICS: Runtime Assurance of Distributed Intelligent Control Systems," 2021 51st Annual IEEE/IFIP International Conference on Dependable Systems and Networks Workshops (DSN-W), 2021, pp. 182-187, doi: 10.1109/DSN-W52860.2021.00038.
