PharmaSight-Act: a physical AI robot that autonomously runs pharma QC and R&D lab tasks

PharmaSight-Act is an autonomous lab robot for pharmaceutical QC and R&D. It is a wheeled humanoid that moves between benches and instruments and carries out hands-on lab work on its own: preparing samples, weighing, diluting, picking vial racks, transferring containers, and operating instruments such as HPLC systems and sonicators.

Pharma labs depend on skilled scientists spending hours on repetitive bench work. That creates bottlenecks, human error and data-integrity risk, and labs face 35-40% workforce attrition. Conventional lab automation is rigid: each fixed workcell does one task and needs re-engineering when the method changes. PharmaSight-Act is built to learn new tasks and work with the existing lab layout and instruments, rather than requiring the lab to be rebuilt around the robot.

The robot runs on BotForge own multimodal robot foundation model, which combines vision, language, depth, touch and action. It has two parts: "The Brain", a vision-language-action model that decides what to do, and "The Imagination", a real-time world model that predicts the outcome of an action before the robot takes it. New tasks are learned in simulated digital twins of the lab and then adapted to the real environment with a small amount of data. Every real-world run feeds back into the data engine, so the system improves with use.

PharmaSight-Act was developed in response to a challenge set by a global top-10 pharma company, which BotForge won, and is now in a development partnership with pfizer. Multiple unit operations demonstrated in simulation (NVIDIA Isaac Sim): opening and closing an HPLC door, picking a vial rack, diluting a chemical, operating a sonicator, transferring containers and handling a weigh boat.

We are moving from simulation to real-world laboratory deployment over the next 12 months, and we want to do it inside real labs. We are looking for: Pilot partners: pharma companies, CDMOs, CROs and QC labs willing to share a defined bench task (for example sample preparation or instrument loading) for a short task study, followed by a digital twin of their lab and a supervised on-site pilot.

• Co-development partners who want a robot trained on their own workflows, with early access ahead of competitors.
• Data and sponsored-research partnerships.
• Seed investors to fund the move to hardware deployment.