Ambi Robotics, formerly Ambidextrous, raises $6.1M for picking robots

Listen to this article

Ambi Robotics, formerly known as Ambidextrous Laboratories, raised $6.1 million in seed funding for its picking robots and operating system that are based on simulation-to-reality artificial intelligence (AI). Co-founded by the great Ken Goldberg, a professor at UC Berkeley, Ambi Robotics said it is exiting stealth mode, although its work on improving robotic grasping hasn’t been a secret in years.

Ambi Robotics has two flagship products. AmbiSort is a robotic putwall that sorts boxes, polybags, and envelopes from bulk input flow (chutes, totes, and bins) into destination containers (mail sacks, totes). Ambi Robotics claims the system works “over 50% faster than manual labor.” AmbiKit is a robotic system that builds unique kits from any item set. The company said it can be used with subscription boxes, medical kits, gift sets and sample sets for a variety of industries, including cosmetics, food and beverage, consumer goods, medical devices, aerospace and automotive.

The company’s robots are modular, but they do use suction-based gripping. Here’s how AmbiSort works. A depth-sensing camera first looks into a bin of items and analyzes the objects. After determining how to best grasp the item, the robot picks up the item with its suction gripper, holds it up to a barcode scanner, then places the item into a bin. The system then alerts a human operator when a bin is full and ready to be packed.

Based on Dex-Net

The company’s robots are powered by its proprietary operating system, AmbiOS, which trains robots in simulation and transfers its advancements to real-world systems. AmbiOS is based off The Dexterity Network (Dex-Net) project that automates the training of deep neural networks to improve a robot’s ability to grasp various items.

Dex-Net was created by researchers at UC Berkeley, many of who are now working at Ambi Robotics. Dex-Net’s algorithms combine the simulation of thousands of 3D object models and synthetic point clouds. Essentially, the deep neural network determines robust pick points that give robots the best chance at successfully grasping novel objects without further training.

Dex-Net 4.0 launched in January 2019 with the ability to train policies for a parallel-jaw and a vacuum-based suction cup gripper on 5 million synthetic depth images, grasps, and rewards generated from heaps of 3D objects. With a physical robot with two grippers, the policy at the time cleared bins of up to 25 novel objects with reliability greater than 95% at a rate of more than 300 mean picks per hour.

“To meet the demand of the staggering growth of online deliveries during COVID-19, our robots work alongside warehouse employees to offset their workload, reduce injuries and improve accuracy, efficiency and throughput,” said Jeff Mahler, co-founder of Ambi Robotics. “Our AI-powered robotic systems are designed for human operation and empower workers to perform at their best, leveraging our simulation-to-reality technology to pick and sort items while workers complete pack-out and handle exceptions.”

Mahler previously told sister publication Robotics Business Review that the Dex-Net team began considering commercializing the technology around 2017 after the 2.0 release. In 2018, the team was invited to give a demonstration of the technology to Amazon, and the enthusiastic and positive response by the e-commerce giant helped inspire the team even more.

“Using our unique approach to deep learning AI based on simulation-to-reality transfer, AmbiOS quickly configures our systems for a variety of sensors, robots and package categories,” said Goldberg. “Our AmbiSort robot and gantry system reliably achieves superhuman sorting – allowing human workers to sort hundreds of thousands of commercial packages at twice the speed of manual picking.”

Warehouses are ripe for automation, and robotics systems are playing catch-up to the growth in e-commerce. There are many robotics companies focused on using manipulation to improve various processes within a supply chain operation. We discussed this during a recent RoboBusiness Direct session “Advances in Robotic Picking, Grasping and Manipulation.” Just yesterday, Boston Dynamics revealed its Stretch robot, a mobile de-palletizing robot designed to unload trucks and move boxes around a warehouse. The key to Strecth’s success will be its ability to handle a variety of packages.