Arsalan Mousavian, senior robotics research scientist at NVIDIA, explains simulation-trained robotics and AI at the August Innovation Forum with host Norman and audience members.
Mousavian describes using NVIDIA Omniverse to train robots in virtual environments, then deploy them in the real world without collecting real-world training data. Simulation can generate realistic scenes, sensor data, and detailed annotations at much lower cost than gathering them manually.
For grasping, the system uses depth images to build 3D point clouds, proposes possible hand positions, and scores their likelihood of success. NVIDIA simulated 17.7 million grasps across 8,800 objects and 262 categories in two days on 16 GPUs. Mousavian reports a 91% real-world success rate without fine-tuning, while noting that retries and harder training scenarios help address failures.
Mousavian outlines a system that detects objects, labels their pixels, and learns a visual signature to match them across video frames. The team generated 350,000 simulated scenes with annotations in one or two days. The technology was combined with grasping and motion planning to find and pick specified objects in a kitchen the model had not been trained on.
For image-based rearrangement, a user provides a picture of the desired setup, and the robot estimates how objects must move to match it. Mousavian also describes turning spoken or written instructions into sequences of robot actions using GPT-3 and existing functions, such as moving fruit to a plate and bottles to a box.
An audience member asks what skills workers will need to operate robots in factories. Mousavian says systems should be usable without advanced robotics expertise, and that manufacturers should communicate their practical needs to technology companies. He points to technology conferences and automation trade shows as ways to connect with potential partners.
Briefs summarise talks and interviews; timestamps open the video at that moment. Claims in a talk are the speaker's own.
Arsalan Mousavian, senior robotics research scientist at NVIDIA, discusses robotics research and manufacturing applications with Norman at the Innovation Forum.
NVIDIA founder and CEO Jensen Huang presents the company’s AI strategy and product announcements at GTC Berlin, with European partners and customers.
NVIDIA CEO Jensen Huang discusses Europe’s AI economy, NVIDIA’s AI infrastructure and agent platforms, and physical AI at GTC Berlin, with examples from European industry and NVIDIA partners.
The host and main speaker from Robotics Central explains Kawasaki’s staged plan to deploy Kaleido humanoid robots, and how Japan’s AI funding, Nvidia computing, and human remote operators fit into that effort.
Amulia Vishban, founder of Techable Ventures, discusses investing in physical AI with Dominic Madori Davis on TechCrunch’s Equity podcast.
Les Carpus, global head of physical AI for NVIDIA’s Inception program, discusses robotics startups and the path to a ChatGPT moment with TechCrunch reporter Tim at Disrupt 2026.
