A lot of the robotics industry is chasing one giant foundation model that could drive any machine through any task. General Robotics is betting against it. Co-founder and CTO Sai Vemprala set out the alternative on Episode 263 of The Robot Report Podcast, published September 25, 2026: split intelligence into specialized pieces and have the system put the right ones together for each job, on whatever robot happens to be doing it.
Vemprala leads development of GRID, the company’s robot intelligence platform. Before co-founding General Robotics he was a senior researcher at Microsoft Research, and he holds a PhD in robotics from Texas A&M University.
Skills, not a monolith
Vemprala laid out the reasoning in a company blog post from September 30, 2025. It describes two paths that general-purpose robotics has swung between. One is brittle, hand-engineered pipelines that work well in narrow domains. The other is large foundation models, which aim broad but need huge amounts of data, are opaque and don’t transfer easily between robots. General Robotics’ answer is modular AI skills: specialized models for perception, planning and control, each solving one part of the problem well. An LLM-based agentic layer reasons about the goal and picks which skills to combine.
In practice, GRID offers a library of composable skills, foundation models and classical techniques that any connected robot can call on. Given a task, the platform:
- works out which skills the task needs and assembles the best combination of models and methods;
- builds and runs the simulation environments needed to train, evaluate and refine a new skill;
- deploys the skill, watches its performance and failures, and decides what to change next.
GRID also uses knowledge graphs to turn each deployment into reusable, structured knowledge. Every robot onboarded, task run, model added and failure logged feeds back into the platform, while customer data and IP stay protected. The company says this improvement “compounds” with every deployment.
From pilot to production
On September 9, 2026, General Robotics announced what it called a major milestone: GRID now “auto-engineers” its own processes, from onboarding new robots and AI models to creating and deploying skills. The company says this sharply cuts the time and specialist robotics expertise needed to put robots into production. The pitch rests on two problems. Engineers who can build intelligent robot systems are scarce, and the robotics stack is so fragmented that every solution needs custom integration.
The platform doesn’t care which robot it runs on. Robot makers in the GRID ecosystem include industrial giant Fanuc, for which General Robotics is an Authorized System Integrator, and Galaxea, which builds bimanual mobile manipulators. According to the company, its customers include global top-5 firms in automotive manufacturing, port operations, energy generation and food and beverage production, plus several government agencies. GRID also integrates NVIDIA Isaac Sim for simulation.
Investors include Construct Capital, Accenture Ventures, E14, Nvidia, Shorooq, Valo Ventures and Khosla Ventures. Accenture announced its investment on April 15, 2026, without disclosing terms. The company is led by CEO and co-founder Ashish Kapoor.
Hans Peter Brondmo, former Google X vice president and CEO of Everyday Robots, summed up the bet: “There’s no single ChatGPT moment coming for robotics, just deployment-by-deployment progress.” For robotics, General Robotics’ view is that better results come from better assembly of many models, not a bigger single one.
