Collaborative Robotics: The Amazon Veteran's Bet on a Robot That Reacts

In June 2022, Brad Porter walked away from a comfortable perch at Scale AI to start something risky. He had spent 13 years at Amazon Robotics building the infrastructure that would eventually put more than 500,000 robots on warehouse floors. He understood, better than almost anyone, what those robots could not do. They could move. They could sort. They could carry. But they could not work with a person, reacting in real time to a shared space.

That gap is what Collaborative Robotics (co.bot) is built to close. The Santa Clara startup's flagship product, Proxie, is a wheeled mobile manipulator that stands roughly human height, rolls on omnidirectional wheels, and uses vision-language models to understand what is happening around it. It is not a humanoid. It is not a simple cart-pusher. Porter describes it as the robot that works beside you, not one that replaces you or requires a roped-off safety zone.

With approximately $140 million raised across three funding rounds, deployments at Mayo Clinic Laboratories and Tampa General Hospital, and an RBR50 Innovation Award in hand, Collaborative Robotics is making a credible case that the cobot era in logistics and healthcare is closer than most operators realize.

AI-generated image

Proxie's Flex Grasp gripper is designed to handle the real-world variety of cart handles, packages, and fixtures found in warehouses and hospitals.

How Proxie Works

Proxie sits in a category sometimes called a collaborative mobile manipulator (CoMM). The base is a wheeled platform. The upper body includes a torso and arms. The whole system is designed to operate in spaces built for humans, without requiring infrastructure changes to floors, aisles, or ceiling fixtures.

The core interaction loop works like this: Proxie receives a task, navigates to the target location using lidar and Visual SLAM, identifies the object or fixture it needs to interact with using onboard cameras and vision models, grasps or hooks the target using its Flex Grasp gripper, then moves with it. If a person steps into its path, Proxie does not stop and wait for a human operator to clear the block. It reroutes, adjusts, or pauses briefly, the way a forklift operator would slow down for a coworker crossing the aisle.

The expressive eyes on Proxie's head unit are not cosmetic. User research in shared-space robotics consistently shows that humans respond better to machines that signal intent. When Proxie's LED eye displays indicate it has seen you and is yielding, workers nearby report higher comfort levels and faster adoption. That kind of trust-building detail is often the difference between a pilot that succeeds and one that quietly gets parked in a corner.

The Zero Charge hot-swappable battery system addresses one of the most underappreciated problems in warehouse robotics: downtime. Robots that must dock for 30-45 minutes to recharge are effectively unavailable for a significant portion of each shift. Proxie's batteries can be swapped by any worker in under a minute, keeping the robot in continuous operation across multi-shift facilities without a dedicated battery management team.

Co.bot unveiled Proxie publicly on November 20, 2024. The timing was deliberate: the company wanted production-ready proof before announcing. By the time the product was public, early pilots were already running.

Under the Hood: Proxie's Technical Architecture

The engineering choices behind Proxie reflect lessons learned from watching a generation of warehouse robots fail at edge cases. Porter's team did not optimize for a single, narrow task. They built a platform designed to generalize across the messy variability of real-world operations.

Core Systems

How Proxie Stacks Up Against Competitors

The NVIDIA partnership is worth flagging here. Collaborative Robotics has been working with NVIDIA's robotics ecosystem, which provides access to Isaac Sim for simulation-based training, CUDA-accelerated inference libraries, and Jetson-class compute reference designs. As vision-language model inference costs drop with each hardware generation, Proxie's reliance on GPU-accelerated onboard compute becomes more cost-competitive for broad commercial deployment.

From Amazon to Co.bot: Brad Porter's Journey

Porter joined Amazon in 2007, two years before Amazon acquired Kiva Systems (later renamed Amazon Robotics) in 2012 for $775 million. That acquisition is now studied in business schools as one of the most consequential robotics bets in retail history. Porter was inside the machine as it scaled from a few hundred robots to more than 500,000 deployed across Amazon's global fulfillment network.

What he saw, repeatedly, was the ceiling. First-generation fulfillment robots are excellent at one thing: moving flat-bottomed pods across flat floors in controlled environments. They cannot handle the irregular geometry of a receiving dock. They cannot navigate a hospital corridor where a gurney might appear. They cannot react to a human reaching past them to grab a package. Every time the task required real-world adaptability, a human had to step in.

After leaving Amazon in 2020, Porter became CTO of Scale AI, the data labeling and AI training company that has become infrastructure for much of the machine learning industry. That stint gave him a different vantage point: not the physical robot, but the data pipelines that make robots smart. He watched foundation models improve at a pace that outstripped most industry forecasts, and he started thinking about what would eventually become Proxie.

He founded Collaborative Robotics in June 2022 with $10 million in seed funding from Bison Ventures. The company deliberately chose Santa Clara over the Boston robotics corridor, positioning itself closer to the AI and semiconductor ecosystem of Silicon Valley. A Seattle office followed the Series B in 2024, putting co.bot within reach of Amazon, Microsoft, and the Pacific Northwest's growing logistics infrastructure.

The investor list tells its own story. Sequoia Capital and Khosla Ventures co-led the Series A. Then Mayo Clinic joined as a strategic investor in that same round, which is rare for an early-stage robotics raise and signals something specific: healthcare was not an afterthought. It was baked into the product roadmap from the beginning. General Catalyst led the $100 million Series B in April 2024, giving co.bot the runway to move from pilots to production deployments.

Where Proxie Is Being Deployed

AI-generated image

Proxie's healthcare deployments focus on transport tasks that consume significant nursing and support staff time, including specimen cart movement between labs and clinical floors.

Mayo Clinic Laboratories

The Mayo Clinic deployment is the flagship case study. Mayo Clinic Laboratories, which processes millions of clinical specimens annually, has been running Proxie in a cart-movement pilot across its facilities. The use case is straightforward on paper but brutal in practice: moving specimen carts between labs, processing stations, and storage areas in an environment where delays have clinical consequences and where the mix of cart types, hallway traffic, and staff movement is constant.

Mayo presented results from the pilot in October 2025. Specific metrics have not been fully disclosed publicly, but the fact that Mayo Clinic, one of the most operationally conservative medical institutions in the world, chose to present the results rather than quietly end the pilot is a meaningful signal. Mayo was also a Series A investor, which means they had early visibility into the technology and chose to bet on it twice before presenting outcomes.

Tampa General Hospital

Tampa General has been running early trials of Proxie alongside the Mayo Clinic work. Hospital logistics is a well-documented problem: nurses and support staff spend a reported 20-30% of their shift time on non-clinical tasks, many of which involve moving things from one place to another. A robot that can handle cart transport without requiring a dedicated human escort is not replacing a nurse. It is giving clinical staff back hours that currently go to logistics that anyone could do.

Warehouse and Logistics

Co.bot has not named all of its warehouse customers publicly, but the company has confirmed deployments with logistics operators alongside the healthcare work. The warehouse use cases center on cart transport, goods movement between zones, and support for picking operations where Proxie moves the cart while a human picker works the shelves. The 1,500-pound tow capacity covers the majority of loaded warehouse carts in common use.

What This Means for the Robot Workforce

The "robots taking jobs" frame is both accurate and incomplete. Yes, Proxie will take tasks. Moving carts between labs is a task. Pushing goods across a warehouse floor is a task. When a robot handles those tasks reliably, the humans who were doing them move to other work, or the facility chooses not to backfill when someone leaves.

The more useful question is what kind of robot Proxie represents relative to what came before. First-generation warehouse robots required significant infrastructure investment: specially marked floors, restricted zones, controlled inventory systems. They worked well only because entire facilities were redesigned around them. Proxie is designed to work in facilities that were not redesigned at all, which changes the addressable market from "large operators with capital for facility renovation" to "any warehouse or hospital that wants a robot next quarter."

The safety profile matters here. Collaborative robots need to pass a higher bar than isolated industrial robots because they share space with people. Co.bot's egocentric sensor suite is designed for this: Proxie is not just detecting obstacles, it is predicting human movement. If a worker's trajectory suggests they are about to step into the robot's path, Proxie adjusts before the conflict happens, not after.

The workforce effects are sector-specific. In healthcare, where the shortage of clinical staff is acute and growing, offloading logistics tasks to robots has a different character than in a warehouse where general labor markets apply. Hospitals are not deploying Proxie to cut headcount. They are deploying it because they cannot hire enough people to do everything that needs doing, and clinical staff time is too valuable to spend on cart transport.

In warehouses, the picture is more mixed. Labor markets in fulfillment have tightened significantly since 2020. Operators who once relied on high turnover and low wages are facing a different reality, and mobile manipulators like Proxie offer a path to predictable throughput without the variability of labor markets. That is a compelling pitch to operations managers who care about consistency as much as cost.

Frequently Asked Questions

The 12-Month Outlook

Collaborative Robotics enters the next 12 months with real deployments, a credible technical platform, and the kind of investor backing that keeps options open. The questions are about scaling, not survival.

The Mayo Clinic results presented in October 2025 are the most important near-term signal. If those metrics land well in broader publication, expect other large hospital systems to accelerate their own pilots. Healthcare robotics moves slowly, but it moves in clusters: once a credible operator publishes positive results, the risk calculus for other operators shifts noticeably.

On the warehouse side, co.bot's challenge is differentiation at scale. Boston Dynamics Stretch is further along in box-handling deployments in distribution centers. Locus Robotics has a larger installed base of goods-to-person AMRs in fulfillment operations. Proxie's differentiator, the ability to work in unstructured environments alongside humans without infrastructure changes, needs to translate into documented ROI cases that procurement teams can put in front of CFOs.

The NVIDIA partnership is a longer-term play. As VLM inference gets cheaper and faster with successive GPU generations, Proxie's AI capabilities become more cost-competitive to operate at scale. Co.bot is building toward a future where the robot's contextual intelligence is a durable competitive moat, not just a current technical advantage that well-funded competitors can replicate quickly.

Porter has spent his career at the intersection of scale and robotics. He knows what it takes to go from pilot to production, and he knows the failure modes. The bet Collaborative Robotics is making is not that robots will eventually work in human environments. That is already happening. The bet is that the transition to truly collaborative, adaptive robots will happen faster than the market currently prices in, and that the company that figures out the human-robot trust problem at scale will own the next decade of the market.