MIT & NVIDIA Researchers Are Building Tech That Could Enable Better Kitchen-Robot Precision

This week, a group of researchers from MIT and NVIDIA are showing off a system that one day may be pivotal in helping our robot chef make dinner without making a mistake.

While robotic planning systems are good at developing high-level plans, they often fail when confronted with complicated scenarios with many obstacles. Because of this, the group wanted to create a task-planning system that performed well in highly-complex environments.

The project focused on developing a task and motion planning (TAMP) algorithm to help robotic systems solve mobile manipulation problems in environments with many articulated and movable obstacles. The core of the algorithm is PIGINet, which the group describes as a transformer-based learning system that, for each proposed task plan, helps the system more quickly and successfully understand the probability of finding motion trajectories associated with the task plan.

Today’s robotic system task planners often fail when faced with the reality of highly complex and infinitely variable real-world scenarios, getting bogged down in processing how to navigate through the unique physical geometries of their environments. The seemingly infinite variety of small things in a kitchen – random items on a counter, the different locations of a pot on a cooktop, open doors and drawers – may be easy for a human to handle but can give a robot fits. With the PIGINet transformer, the system will be able to more quickly process through and understand the success probabilities of each course of action due to the specific start state and the given obstacles within.

According to the group, the PIGINet transformer-enabled task planner gives the robot a better chance of success by better understanding the various scenarios and each’s feasibility before they are executed. Their initial experiments showed that using PIGINet substantially improves planning efficiency, cutting down runtime by 80% on problems in relatively simple scenarios and up to 50% in more complex ones.

While the group’s initial effort focused on kitchen and food-planning tasks, it believes its system can be applied to other tasks within and outside the home.

While there have been a lot of venture capital dollars and product development hours spent on developing kitchen automation, you can see by this project and those similar to it just how early we are in developing truly advanced kitchen automation. The kitchen is one of the most complex and variable work environments, and creating a robot that doesn’t simply automate a single repeatable process is extremely difficult.

Here’s hoping that this project and ones like EPIC Kitchens are laying the foundation for the robot chefs of the future.

You can watch a video on their project and how it works below:

Source link

This post originally appeared on TechToday.