Covariant Unveils AI-Powered Warehouse Robots

Pieter Abbeel’s new robotics startup Covariant this week deployed their AI-equipped robot at customer facilities in North America and Europe in the apparel, pharmaceutical, and electronics industries. 

The company is backed by some of the biggest names in AI, including Geoffrey Hinton, Jeff Dean, Yann LeCun, Fei-Fei Li, and many others.  

Their GPU-accelerated platform consists of off-the-shelf robot arms equipped with cameras, a special gripper, and of course, GPUs. 

“Even though we are just getting started, the systems we have deployed in Europe and North America are already learning from one another and improving every day,” said Pieter Abbeel, founder, President and Chief Scientist, in a recent press release

Abbeel’s lab at UC Berkeley has pioneered many of the AI breakthroughs seen in autonomous machines today. One of his lab’s research areas is the development of reinforcement learning models, in which an algorithm trains over and over through trial and error until it learns how to complete a task successfully with reinforcement. 

Covariant-powered robot at Obeta

In the video, Covariant’s robot is seen at German logistics company Obeta, demonstrating how it picks up items from storage totes and adds them to individual order boxes for shipping. 

In the backend, the company’s deep learning models are trained using NVIDIA GeForce 2080Ti GPUs, as well as NVIDIA Quadro RTX 6000 GPUs

The company says they use the cuDNN-accelerated, deep learning PyTorch framework as their go-to platform.

For inference, the robots rely on NVIDIA Quadro RTX 6000 GPUs, which process the robot’s sensory inputs, perform object detection, and determine the actions that the robotic arm should take. At the German factory, Covariant’s AI-based system can detect over 10,000 different items with 99% accuracy. 

Covariant founders Tianhao Zhang, Rocky Duan, Peter Chen, Pieter Abbeel

“Covariant robots learn general abilities such as robust 3D perception, physical affordances of objects, few-shot learning and real-time motion planning, which enables them to quickly learn to manipulate objects without being told what to do,” the company stated in a press release. 

The company says they plan to build more robots in industrial-scale settings, including manufacturing, agriculture, hospitality, commercial kitchens, and eventually people’s homes. 

(opens in a new tab)”>Read more>

Discuss (0)