Tutor Intelligence Releases Second-Generation Cassie and Sonny Warehouse Robots

en.Wedoany.com Reported - On September 16, U.S. robotics company Tutor Intelligence released its second-generation Cassie and Sonny warehouse robots. Both products use the company's self-developed robot foundation models and are designed for warehousing, manufacturing, and supply chain operations. The second-generation Cassie has been upgraded from a fixed single-arm robot to an autonomously mobile platform capable of performing tasks such as depalletizing, loading, palletizing, repacking, and full-case picking in warehouses and factories; Sonny adopts a dual-arm semi-humanoid structure and focuses on scenarios requiring greater operational flexibility, such as shelf picking, e-commerce fulfillment, assembly, and packaging.

Cassie is equipped with Tutor's Turing 200-million-parameter robot foundation model, perceiving its environment through vision and executing new tasks. The company disclosed that Cassie has already been put into production use in multiple factories and warehouses in the United States, completing millions of large-item and full-case picks cumulatively. The second-generation model adds a four-wheel autonomous mobile chassis, with a battery life covering more than one work shift, and also integrates an onboard vacuum generation system, eliminating the need to install additional external vacuum equipment for fixed workstations.

Safety and navigation systems have also been upgraded simultaneously. The second-generation Cassie adds two sets of 3D LiDAR on top of the original ISO-certified collaborative hardware and 2D LiDAR, used for personnel detection, obstacle recognition, and collision-free autonomous navigation. The company plans to deploy the new-generation Cassie for the first time in an end-to-end full-case picking process in the fall of 2026, enabling it to move autonomously between different workstations and directly handle case transport and picking tasks in B2B order fulfillment.

Sonny, meanwhile, has shifted from a previous research platform to a deployable industrial robot. The new-generation Sonny will become Tutor's first commercial robot equipped with the Ti0 vision-language-action model. Ti0 has 4.5 billion parameters and is mainly used for warehousing tasks requiring dual-arm coordination, complex object recognition, and multi-step operations. The first deployment scenario is e-commerce warehouses, where the robot can autonomously move within existing warehousing environments, grasp different types of goods from shelves, and deliver completed orders to packaging stations.

To train Sonny and the Ti0 model, Tutor has built Data Factory 1, a robot data factory. The system consists of 100 Sonny robots, with remote operators in different regions demonstrating new tasks through VR teleoperation and manually correcting the robot when errors occur. The collected data is used for post-training stages such as behavior cloning, supervised fine-tuning, and reinforcement learning.

Data Factory 1 also employs a parallel robot training mechanism. The same policy can run simultaneously on 100 robots. The company disclosed that edge cases that might originally require a single robot to run for about 8 hours to encounter once can appear in about 5 minutes when 100 robots run in parallel. After each task round, operators give positive or negative ratings to the robot's actions, and the relevant feedback continues to be used for model reward training and behavior optimization.

Tutor stated that Cassie and Sonny currently use different foundation models: Cassie runs the Turing series models, while Sonny runs the Ti0 model. The two types of robots correspond to different warehousing tasks, with Cassie mainly handling larger goods such as pallets and cartons, and Sonny mainly handling individual items and tasks requiring fine dual-arm manipulation. The company's current robot deployment scale is growing by about 20% per month, and it is expanding its robot manufacturing capacity in the United States.