Berkeley and NVIDIA Unveil Agentic Robotics at CoRL 2026
Overview
At the 2026 Conference on Robot Learning, UC Berkeley and NVIDIA announced 'agentic robotics,' a method using LLMs to generate robot control code directly from natural language. The system decomposes tasks into specialized coding agents that iteratively refine programs in simulation before deployment, aiming to overcome data scaling limits.
Generated from attributed reports · Updated 11 hours ago
Event evidence and corrections
1 attributed source owners. Ownership does not establish independent confirmation. Quantities are reported separately and are never added together.
Reported quantity · Reported success rate: 99 percent · reported trials · Differing source assertions
“er agent tries to rip it apart and find flaws.” In UC Berkeley’s testing , agentic robotics approaches were able to reach success rates of up to 99 percent, matching human hand-engineered performance, even in tasks where baseline imitation learning techniques saw success rates of only 20 percent. But as Goldberg points out, agentic robotics is not just about performance, but also about efficiency and safety. “You’re”
Exact source · revision 2Source owner not reported
Reported quantity · Reported success rate: 20 percent · reported trials · Differing source assertions
“er agent tries to rip it apart and find flaws.” In UC Berkeley’s testing , agentic robotics approaches were able to reach success rates of up to 99 percent, matching human hand-engineered performance, even in tasks where baseline imitation learning techniques saw success rates of only 20 percent. But as Goldberg points out, agentic robotics is not just about performance, but also about efficiency and safety. “You’re”
Exact source · revision 2Source owner not reported
Environment: simulation
“next. Ken Goldberg, a UC Berkeley roboticist and professor of engineering, said new models make it possible for teams of AI agents to develop robot-control programs, test them in simulated environments and revise the code when something doesn't work. Goldberg, Spencer Huang, director of robotics engineering at NVIDIA, and a team of researchers call the new approach "agentic robotics." They said it could offer a way”
Exact source · revision 2science-x
Reported quantity · packages sorted annually per robot: 15000 other · Basis not reported
“Berkeley-based Ambi Robotics, which specializes in logistics and warehouse order fulfillment, applied the new methods earlier this year to analyze parcel-sorting data and test updates to robot behavior. Tests showed an increased capacity of over 15,000 packages sorted annually per robot; the change is now deployed across 30% of the company's U.S. fleet. "AI-powered industrial robots must continuously improve reliabi”
Exact source · revision 2science-x
Reported quantity · robots in fleet: 30 other · reported trials
“Berkeley-based Ambi Robotics, which specializes in logistics and warehouse order fulfillment, applied the new methods earlier this year to analyze parcel-sorting data and test updates to robot behavior. Tests showed an increased capacity of over 15,000 packages sorted annually per robot; the change is now deployed across 30% of the company's U.S. fleet. "AI-powered industrial robots must continuously improve reliabi”
Exact source · revision 2science-x
Reported quantity · Reported trials: 1 other · reported trials
“created and fine-tuned a physics simulation that accurately reproduced the video, including very realistic deformations of the sponge. The researchers then gave the AI that simulation and their Graph-as-Policy framework. The AI generated a robot-control program and repeatedly tested and improved it in simulation. When the researchers ran the resulting code on a real robot, it worked. In one trial, Goldberg said, the”
Exact source · revision 2science-x
Artifact availability · code: available
“icrobiology lab background with pharmaceutical news experience in oil, gas, and renewable industries. Full profile → Robert Egan Bachelor's in mathematical biology, Master's in creative writing. Well-traveled with unique perspectives on science and language. Full profile → Citation : AI agents test and improve robot code in virtual worlds before deployment (2026, November 9) retrieved 9 November 2026 from https://tec”
Exact source · revision 2science-x
Developments
- 2026-11-09 14:47 UTC · 1 reportsCan Agentic Robotics Solve Robotics' Data Scaling Problem?Ken Goldberg — X Robotics：Lucid report from @BotJunkie: Can Agentic Robotics Solve Robotics’ Data Scaling Problem? https://spectrum.ieee.org/agentic-robotics
- 2026-11-09 14:00 UTC · 2 reportsAgentic Robotics Approach for Industrial TasksIEEE Spectrum — Robotics：Can Agentic Robotics Solve Robotics' Data Scaling Problem?
Report timeline
Follow attributed reports and material updates.
- Ken Goldberg — X RoboticsLucid report from @BotJunkie: Can Agentic Robotics Solve Robotics’ Data Scaling Problem? https://spectrum.ieee.org/agentic-robotics
Lucid report from @BotJunkie: Can Agentic Robotics Solve Robotics’ Data Scaling Problem? https://spectrum.ieee.org/agentic-robotics
- Tech Xplore — RoboticsSignalAI agents test and improve robot code in virtual worlds before deployment
Ken Goldberg of UC Berkeley, Spencer Huang of NVIDIA, and collaborators present 'agentic robotics,' a method where teams of AI agents generate, test, and refine robot-control programs in simulated environments using NVIDIA's Isaac framework and an open Graph-as-Policy architecture.
- IEEE Spectrum — RoboticsSignalCan Agentic Robotics Solve Robotics' Data Scaling Problem?
At the 2026 Conference on Robot Learning, UC Berkeley and NVIDIA announced agentic robotics, a method using large language models to write robot control code directly from natural language prompts. This approach aims to overcome the plateau of imitation learning by treating robot programming as a coding problem rather than a data collection one. The system decomposes tasks into sub-tasks handled by specialized coding agents (e.g., perception, manipulation) and iteratively refines code in simulation, even adjusting simulation parameters to match real-world video. While testing showed up to 99 percent success rates matching human-engineered performance on specific tasks, the authors acknowledge challenges with sim-to-real physics gaps and AI-generated code reliability.
Event coverage history
Current attention 22·Peak within the comparable range 29（2026-11-09 15:00 UTC）·Change within the comparable range over 24 hours –
The trend compares participants observed continuously; its scope may be narrower than current attention totals. Hover or click to explore hourly observations, or use the left and right arrow keys.
Timezone · UTC
Article dates follow your selected timezone. Briefing editions use Hong Kong time (UTC+8).