Tutor Intelligence launches second-generation intelligent warehouse robotics with a classroom to teach them

Tutor Intelligence Inc. has unveiled its latest advancements in intelligent warehouse robotics with the introduction of the second-generation Cassie and Sonny robots, powered by the company’s advanced robot foundation models. Sonny, initially developed as a semi-humanoid robot, has evolved into a fully deployable robot equipped with autonomous manipulation and mobility abilities within warehouses.

Cassie, a large-format, single-arm robot, has been enhanced to facilitate mobility in warehouse and factory settings, with improved pallet movement and physical intelligence.

Tutor Intelligence co-founder and CEO Josh Gruenstein emphasizes that the company aims to create robots that are not only generally capable but also immediately useful in practical applications. They recognized the current dominance of humanoid robots in the industry but identified a need for robots designed around specific tasks rather than mimicking human appearance.

Cassie and Sonny represent two distinct approaches to this challenge: Cassie provides the muscle for moving heavy materials in expansive pallet warehouses, while Sonny excels in environments with shelves where arm and hand functionality is essential for picking and placing small objects.

Sonny, originally designed as a dual-armed semi-humanoid robot, has undergone significant development since its introduction alongside Tutor’s Ti0 4.5 billion parameter Vision Language Action model. The robot has transitioned from a research platform to an autonomous entity capable of manipulating and moving around in warehouses. Cassie, on the other hand, has been adapted for warehouse mobility, incorporating pallet handling capabilities and enhanced physical intelligence.

In an innovative approach to training these robots, Tutor Intelligence has established a "classroom" for AI model training called Data Factory 1. This facility houses 100 Sonny robots that engage in a learning process referred to as "tutoring." Experts remotely control the robots to demonstrate tasks such as object identification, grasping, movement, and placement, thereby providing real-world visual and kinetic data essential for AI model pre-training.

As Cassie does not require extensive reasoning about a wide range of potential scenarios, its physical vocabulary is relatively simple, involving basic actions like identifying, grasping, moving, and placing objects. Sonny, with a larger vocabulary, necessitates the development of a powerful AI model, Ti0, to handle its more complex tasks.

Tutor Intelligence has assembled a team of "tutors" who can remotely instruct robots on new tasks, and once trained, the 100 robots in the classroom undergo post-training to refine their behaviors and ensure smooth operation in actual warehouse environments. This iterative process involves reinforcement learning, where successful actions receive positive reinforcement (a thumbs up), and unsuccessful ones receive corrective feedback (a thumbs down).

This mechanism enables the AI models to continuously improve and surpass human performance in specific robotic tasks.