VR, AI Help Bots Collaborate with Humans

Article By : Ann R. Thryft, EE Times

MIT's VR system lets workers teleoperate robots, while Kindred Systems and Veo Robotics are using AI to create collaborative robots that can safely and efficiently work alongside humans.

BOULDER CREEK, Calif. — Robotics technologies that can help factory and warehouse workers do their jobs better, more safely, and even remotely are emerging from research labs and startups. A team at the Massachusetts Institute of Technology has developed and tested a virtual-reality system that lets workers teleoperate robots without the lag time or user side-effects that have hampered other VR-based approaches. Startups Kindred Systems and Veo Robotics, meanwhile, are using artificial intelligence (AI) to create robots that can work collaboratively alongside humans.

To let factory workers telecommute — at first glance a no-op kind of concept — MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) has developed a system that integrates commercial VR technology with existing robotics control software to enable a virtual shop floor. The engineering team, led by CSAIL director Daniela Rus, took an approach in between two traditional methods for teleoperation via VR: the direct and cyber-physical models.

The direct model couples the user’s vision directly to the robot’s state. This limits the user’s viewpoint to only one perspective and can cause headaches and nausea when signals are delayed. In the cyber-physical model, the user interacts with a virtual copy of the robot and its environment, but the approach requires a lot more data and makes it difficult to respond to constantly changing situations.

CSAIL’s system integrates commercial VR technology with existing robotics control software. To provide the sense of collocation, the system maps the user space into the virtual space and then maps the virtual space into the robot space. Commercial VR game engines render most of the environment. Users wear a headset in a VR control room with multiple sensor displays. They use the headset’s controllers to interact with controls that appear in the virtual space to complete tasks, such as opening and closing the robot arm’s hand grippers to pick up items and stack them for assembly. 

 

Combining a VR gaming headset and hand controllers with existing robotics control software, CSAIL's new VR system lets users teleoperate a factory robot from up to hundreds of miles away. Source: MIT Computer Science and Artificial Intelligence Lab

Combining a VR gaming headset and hand controllers with existing
robotics control software, CSAIL’s new VR system lets users
teleoperate a factory robot from up to hundreds of miles away.
Source: MIT Computer Science and Artificial Intelligence Lab

Users receive continual visual feedback from the virtual environment via stereo cameras on the robot’s head, so there’s no signal delay. Instead of extracting 2-D information from each camera, building a 3-D model of the physical environment, then processing and redisplaying the data, the CSAIL system lets the user’s brain extract 3-D information from the stereo camera images. It requires fewer resources than traditional approaches and thus lowers costs, according to MIT.

The CSAIL team developed the system using an Oculus Rift headset and Rethink Robotics’ Baxter robot, but the technology can work with the HTC Vive headset and other robot platforms. In tests, the CSAIL system was better at grasping objects 95 percent of the time, and 57 percent faster at doing tasks, than a state-of-the-art system, the team said. The developers also demonstrated that users can control the robot from hundreds of miles away.

AI assist for robotic coworkers
Startup Kindred Systems said it has begun pilot programs with several global retailers for Kindred Sort
, an AI-powered robotic solution that helps retail distribution and fulfillment-center workers assemble orders from batch-picked items. Kindred’s software lets robot arms pick up, barcode scan, and sort multiple sizes and shapes of items, at a rate of 250 to 400 times an hour. Gap Inc. is one trial customer, a Fortune magazine article reports. The robot arms are made by Fanuc, workers manipulate them using videogame-like moves, and the software uses deep-learning and reinforcement-learning techniques, according to the article.

AI-powered Kindred Sort software lets robot arms pick up, barcode scan, and sort multiple sizes and shapes of items to help workers in retail fulfillment centers. Source: Kindred Systems

AI-powered Kindred Sort software lets robot arms pick up,
barcode scan, and sort multiple sizes and shapes of items
to help workers in retail fulfillment centers.
Source: Kindred Systems

Veo Robotics said it is combining AI, advanced computer vision, and 3-D sensing to make traditional industrial robots in workcells safer for people to be around. Its first product will use “a new class of intelligent algorithms running on parallel-computing hardware and distributed 3-D sensors” to make big, powerful industrial robots capable of close collaboration with humans, according to a statement. Key vertical markets include automotive, consumer packaged-goods, and household appliance manufacturers, as well as automated distribution centers.

— Ann R. Thryft is the Industrial Control & Automation Designline editor at EE Times.

Subscribe to Newsletter

Leave a comment