Dexmal MaaS Upgrades to Shift Robotics from Custom Projects to Scalable Daily Active Usage

This analysis explores the economic model of embodied AI, contrasting API-based model access with robot daily active usage (DAR) as a key metric for commercial viability. It examines the challenges of cross-platform generalization in physical robotics and details how the Dexmal MaaS platform aims to lower barriers for data processing, fine-tuning, and hardware integration. The piece argues that true value lies not just in model calls but in robots consistently performing tasks in real-world scenarios like logistics and manufacturing.

Physical Intelligence has upgraded its Dexmal MaaS platform to lower barriers for post-training and robot integration, offering standardized pipelines like SFT, DAgger, and RECAP alongside data processing tools. The service supports multiple robot bodies via DexOS, aiming to shift the industry from custom project delivery toward a scalable model where developers handle specific applications while model providers offer foundational capabilities. Success is measured by daily active robots performing stable tasks in real scenarios rather than just deployment counts.