Household Robot Navigation System Developed by MIT Researchers

MIT researchers have developed a neural network program bringing “task and motion planning” to domestic robots. 

The new platform, dubbed PIGINet (Plans, Images, Goal and Initial facts), uses simulated home environments to map a space and build movement plans for domestic robots as they perform different tasks.

Manufacturing and industrial robots have seen increased uptake due to the relatively fixed nature of facilities’ layouts. By contrast, home spaces are often less predictable, with robotic navigation posing a challenge in these environments. MIT’s system is designed to meet the current gap in the market for home robots, providing a means to ensure the robot’s navigation through the space.

According to the research team, the system uses a learning-enabled Task and Motion Planning (TAMP) algorithm that adapts a robot’s movements to meet the challenge of movable obstacles.

“The method … takes in a task plan, the goal and the initial state, and predicts the probability of finding motion trajectories associated with the task plan,” the team said. “We integrate PIGINet within a TAMP planner that generates a diverse set of high-level task plans, sorts them by their predicted likelihood of feasibility and refines them in that order.”

The team said in experiments where PIGINet was trained on 150 to 600 problems, the platform reduced runtime by 80% in small spaces and by 10%-50% in larger areas.

While the model is currently focused on kitchen spaces, the team said they hope to expand to other rooms in the house, as well as one day outside of the domestic space.