Hedra’s new Model Factory cranks up the production volume for physical AI model training

Hedra Inc., a startup specializing in autonomous artificial intelligence agents, has unveiled its Hedra Model Factory – a post-training and inference platform designed to streamline the development of physical AI models for real-world applications. The platform, which pairs Hedra's research teams with AI agents to orchestrate training runs, data curation, and autonomous iteration, aims to address the significant data and compute costs associated with training models for robots, manufacturing, biology, and defense workloads.

Hedra's Chief Executive Michael Lingelbach, a former reinforcement learning researcher at Stanford University and Nvidia Corp., believes that physical AI is the next major frontier in AI development. The Model Factory offers an end-to-end production platform for physical AI models, allowing customers to provide base models and proprietary simulation and video data, which is then processed within a flexible graphics processing unit sandbox environment powered by Hedra's proprietary inference engine.

The RSI process, a relatively new technique in AI, enables the autonomous evaluation, training, and improvement of model architectures. This continuous loop optimizes models for specific tasks, with vision-language models scoring data and providing live reinforcement learning feedback.

Hedra has demonstrated the Model Factory's capabilities by retraining MiniMax Group Inc.'s H3 video model and distilling the Krea 3 2K image generation model, achieving significant improvements in performance and speed. The startup's ultimate goal is to create thousands of cost-effective, bespoke models to usher in the physical AI revolution by optimizing training as much as possible.

With the Hedra Model Factory service now available, enterprises can quickly develop specialized autonomous systems to bring intelligence into real-world environments.