Seeking to answer the question posed about the behavior of domestic robots, MIT's Computer Science and Artificial Intelligence Laboratory (CSAIL) has presented a system called PIGINet, which aims to improve the problem-solving capabilities of domestic robots using machine learning.
Since they tend to follow predefined orders to perform tasks, and therefore fail to be efficient in changing environments, the new PIGINet system eliminates unfeasible task plans and significantly reduces planning time.
With the evaluations performed, it was detected that the robot that tries several task plans to decide which one to perform, takes a long time to perform the requested task, being inefficient and time consuming at the end, this new system, PIGINet, improves this action and reduces the planning time.
To test this system, the team created simulated environments with different tasks, the results showed that PIGINet reduced the planning time in different scenarios, up to 80% in simpler scenarios and between 20% and 50% in more complex scenarios.
PIGINet is not only limited to homes, but it also has practical applications in other environments. Researchers continue to refine the system to suggest alternative task plans and accelerate the generation of feasible task plans without the need for large data sets.
MIT notes that, artificial intelligence experts have praised PIGINet's approach to improving the efficiency of domestic robots. Leslie Pack Kaelbling, a researcher on this project, states that this system offers reliable and efficient solutions for a wide range of situations.
According to Zhutian Yang, a PhD student at CSAIL and lead author of the paper, the goal is to optimize PIGINet to provide more efficient and adaptable domestic robots. This approach could revolutionize the way robots are trained and applied in the homes of all users. To learn more about this breakthrough, visit:
22 de Agosto, 2023