Microagi hires Animesh Garg to train robot models for factory jobs
The Georgia Tech professor joins as chief research officer with seven researchers, bringing a lab-built startup into Microagi's push to deploy robots in factories.
By RuntimeWire Staff · Published
Primary source: Microagi on X
Why it matters
Microagi is investing in the layer between general robot capability and reliable factory work. Garg's team brings research depth to that problem, while real customer deployments will determine whether task-specific adaptation can scale beyond pilots.
On September 29th, Animesh Garg (@animesh_garg) joined Munich robotics company Microagi as chief research officer, bringing seven members of his research group and a stealth startup they had been building. Garg will lead the work of adapting general-purpose robotics models to particular jobs and factory settings, Microagi said in its announcement. The appointment brings a research program into Microagi's effort to put robots to work on existing industrial lines.
Garg has spent his career moving between robotics research and deployment. He worked at NVIDIA Research for six years, served as chief scientific officer at humanoid maker Apptronik from 2024 to 2025, and became a Georgia Tech professor in 2024. At NVIDIA, his simulation research contributed to ORBIT, which later became part of the company's Isaac Lab robotics-learning framework, according to Business Insider's report. His academic training spans manufacturing processes and automation at the University of Delhi, industrial engineering at Georgia Tech, and computer science and operations research at UC Berkeley, followed by a postdoctoral appointment at Stanford AI Lab, according to Garg's biography.
Garg described the commercial moment as the reason to make the move. "The technology has crossed a threshold," he told Business Insider. "We have gotten to a point where we can create sustainable businesses." Microagi's announcement says the decision followed an 11-week collaboration between Garg's group and Microagi's researchers, before the two sides discussed terms.
The work after the model
Microagi calls Garg's focus post-training: taking a broadly capable robot model and adapting it until it can perform one task reliably in one operating environment. The phrase is borrowed from AI model development, where post-training shapes a pretrained model for practical use. In robotics, the target is physical repetition: a machine completing the same factory job through changing shifts and conditions, without requiring an engineer to supervise each run.
That is a specific commercial bet. General-purpose robot models aim to transfer across tasks; industrial customers need performance on the task in front of them. Microagi is combining that adaptation work with Atlas, which deploys robotics on existing hardware and adapts models to customer operations, and Shift, which records physical work as training data. Microagi says Shift collects first-person data across 15 countries. Its announcement says the incoming group will work with multimodal information from simulation and real environments, including footage from workers wearing cameras and data from sensor-equipped gloves.
The strategy gives Garg's research a direct route to production settings, while Microagi adds an experienced robotics-learning group to the part of its business that connects data to task-specific performance. Microagi says it acquired the stealth startup Garg's group had been building to bring the team in. The terms were not disclosed, and the seven researchers were not individually named in the announcement.
Garg's move also reflects a broader change in where robotics research gets commercialized. The Business Insider report described other academics moving into startups, including Carnegie Mellon professors Deepak Pathak and Abhinav Gupta, who cofounded Skild AI, and Stanford professor Chelsea Finn and UC Berkeley professor Sergey Levine, who cofounded Physical Intelligence. Those companies are pursuing general-purpose robot models. Microagi is placing its emphasis further down the path to deployment: adapting models and data to a particular customer's work.
Capital for deployment
Microagi has financing to pursue that work. On July 16th, it announced a $55 million seed round led by Hummingbird Ventures, with Northzone, LocalGlobe, Village Global and redalpine participating. The deal, which RuntimeWire covered at the time, was described by Microagi and its backers as the largest seed round raised by a German startup. The valuation was not disclosed.
The appointment adds a research organization to Microagi, which has already raised capital to build its data-collection and deployment operation. Microagi's long-term target is to deploy one million robots in the real world; CEO Bercan Kilic told Business Insider Microagi wants to deploy at least 10 million robots for customers over five years. Both figures are targets set by Microagi, not deployment totals.
For Garg, the move puts his research thesis closer to the conditions it has to survive: specific jobs, customer operations, and the uneven data available outside a lab. For Microagi, the test is whether post-training can make deployment repeatable across sites and tasks. The hiring brings expertise and a team; the transaction terms, customer-level performance and scale of deployed robots remain outside the announcement's evidence.