New AI-based approaches for controlling autonomous robots
Tom Cruise’s character Maverick is tasked with teaching young pilots how to accomplish a seemingly impossible task. They must fly their jets into a canyon so deep that they are not detected by radar and then climb out at a steep angle to avoid the walls of the canyon. Spoiler alert! With Maverick’s help, the human pilots complete their mission.
Machines, however, would find it difficult to accomplish the same task. For an autonomous aircraft for example, the easiest path to the target may conflict with what it needs to do in order to avoid collision with canyon walls or remain undetected. The stabilize-avoid conflict is a major problem for many AI methods. They are unable to solve it and reach their goals safely.
Researchers at MIT have developed a technique that solves complex stabilize-avoid issues better than any other method. The machine-learning method matches or exceeds existing methods in terms of safety, while providing a 10 fold increase in stability. This means the agent is able to reach and remain stable within the goal region.
Source:
https://techxplore.com/news/2023-06-ai-based-approach-autonomous-robots.html