Cobot Announces Second-Generation Proxie, Bringing Production-Tested Physical AI to Real Operations

Collaborative Robotics Inc. (Cobot) today announced the second-generation Proxie, a general-purpose mobile collaborative robot that identifies the work that needs to be done and does it, with zero software integration.

Generation 2 is built on the learnings from the first generation Proxie that has now logged 12,627 operating hours in production environments ranging from hospitals to manufacturing to logistics. Often operating 16-hour days, in the past 2 years of operations, Proxie has moved over 40 million pounds of materials and saved over 17 million steps.

"For decades, deploying robots has meant choosing between mobility and dexterity, and always required custom software integration," said Brad Porter, founder and CEO of Cobot. "Our second-generation Proxie brings all of that together in one platform we designed end-to-end."

In production, Proxie builds a real-time world-model (RTWM) of the operations which it then uses to self-identify work that needs to be performed and completes the work autonomously. Cobot calls this capability Autotasking, and it runs with no software integration and no human dispatchers.

With over 500 improvements derived from Proxie's experience in the field, Generation 2 is designed for scale-up manufacturing with simplified assembly, 40% fewer parts, and lifecycle-tested components. Gen2 is more compact, saving space in narrow hallways and elevators, while packing more strength to move up to 1,500lb carts and lift up to 200lbs on its vertical spine.

Proxie Gen2 moves to lithium iron phosphate batteries for industrial safety and a self-swapping battery station for continuous operation with no downtime.

Proxie Gen2 also introduces a modular option for two-armed interactions. Leveraging the latest in physical AI models for dexterous bimanual manipulation, Proxie can quickly be trained to complete complex manipulation tasks reliably.

"For decades, deploying robots has meant choosing between mobility and dexterity, and always required custom software integration," said Brad Porter, founder and CEO of Cobot. "Our second-generation Proxie brings all of that together in one platform we designed end-to-end. It moves, manipulates, and orchestrates its own work. The robots identify what needs to be done, announce what they're going to do, and then they just do it."

Physical AI on the Robot

With its on-robot AI compute, task inference runs locally on Proxie, with no cloud dependency for core task execution. Proxie's ScoutSense sensor technology sees the workplace from human eye level, plans and sequences the work, and audibly announces each action before the robot moves. The new bimanual configuration introduces two arms able to cover a wide array of human-scale, two-handed tasks including hospital resupply, warehouse kitting, life sciences lab operations, and manufacturing line tending.

Vista: Orchestration and operational intelligence across the fleet

Cobot's Vista AI console gives operators real-time visibility into every task as it is created, queued, and completed, tracks fleet status, and surfaces operational patterns so teams can intervene before issues affect a shift. Where Business Intelligence (BI) tools explain what already happened, Vista manages what is happening now and helps shape what happens next.

Built with NVIDIA and AWS

Cobot has expanded its collaborations with NVIDIA and AWS. The NVIDIA Jetson platform provides on-robot AI compute for real-time perception, planning, and local task inference, letting Proxie execute core work on the edge. Cobot is also collaborating with NVIDIA Robotics on NVIDIA Isaac Sim and NVIDIA Omniverse NuRec for simulation, turning real-world deployment data into repeatable scenarios that accelerate validation and rollout across customer environments.

AWS serves as the cloud backbone for Cobot's fleet-level intelligence, powering Vista and Autotasking, enabling the model training behind Proxie, and turning individual robots into a continuously improving autonomous fleet.

"Physical AI is moving into real-world operations, where robots need to understand dynamic environments, reason in real time, and work safely alongside people without months of custom integration," said Amit Goel, head of robotics and edge AI ecosystem at NVIDIA. "Cobot's integration of NVIDIA Jetson for on-robot AI compute and NVIDIA Isaac Sim for simulation and validation helps Proxie move from development to deployment faster, bringing more adaptable automation to hospitals, logistics and manufacturing environments."

Availability

The second-generation Proxie is available to order now starting at $5,000 per month. Cobot will demonstrate the platform daily at Automate 2026, June 22-25, at McCormick Place in Chicago. Cobot will be in booth #1684 in the Humanoid Pavilion, hosted by NVIDIA.

About Cobot

Founded by renowned robotics leader Brad Porter, Cobot builds robots that take on the work the world can no longer staff. Through our first robot, Proxie, we are bringing physical AI into hospitals, warehouses, labs, and factories where work actually happens. Proxie is in production today with enterprise customers across healthcare, manufacturing, and logistics. Cobot is headquartered in Santa Clara, Calif. Learn more at co.bot.