Listen to this article now

Why aren’t there more robots in homes? This is a surprisingly complex question, and the answer lies in the intricate nature of our homes. While autonomous systems have found success in warehouse and factory settings, their adoption in homes has been limited. One of the main reasons for this is the relative ease of navigating structured environments found in warehouses and factories. Although these systems typically require an initial mapping of the space, once that is done, there is usually little variation in the environment.

Homes, on the other hand, present a multitude of challenges. They vary significantly from one unit to another, and they are filled with obstacles that are not robot-friendly. Furthermore, homes tend to be dynamic environments where furniture is often moved around, and objects are left on the floor. While robotic vacuums have become popular in households, they are still being improved after decades on the market.

Addressing this issue, researchers at MIT CSAIL have developed PIGINet (Plans, Images, Goal, and Initial facts) to enhance the capabilities of home robotic systems in task and motion planning. PIGINet is a neural network designed to streamline the creation of action plans in different environments.

According to MIT, PIGINet employs a transformer encoder, a versatile and state-of-the-art model designed to process data sequences. In this case, the input sequence consists of information about the task plan, images of the environment, and symbolic encodings of the initial state and desired goal. By combining these inputs, the encoder generates predictions about the feasibility of the selected task plan.

Currently, PIGINet primarily focuses on kitchen-based activities. It leverages simulated home environments to create plans that involve interactions with various elements such as counters, cabinets, the fridge, and sinks. The researchers report that in simpler scenarios, PIGINet reduced planning time by 80%. For more complex situations, the reduction ranged from 20% to 50%.

The MIT CSAIL team believes that PIGINet has broader applications beyond just households. Zhutian Yang, a PhD student involved in the project, states, “The practical applications of PIGINet are not confined to households. Our future aim is to further refine PIGINet to suggest alternate task plans when infeasible actions are identified. This will accelerate the generation of feasible task plans without the need for extensive training datasets to develop a general-purpose planner from scratch. We believe that this advancement could revolutionize the way robots are trained and applied in homes.”

In summary, the complexity and variability of home environments have posed challenges to the widespread adoption of robots. However, through innovations like PIGINet, researchers are making significant progress in enabling robots to navigate and perform tasks in these dynamic settings.