Yuanli Lingji Upgrades Dexmal MaaS with Post-Training Capabilities for Embodied AI Models

Yuanli Lingji has upgraded its Dexmal MaaS platform to offer post-training services for embodied AI models, enabling developers to upload robot operation data for cloud-based training and deployment. The upgrade supports three pipelines: Supervised Fine-Tuning (SFT), DAgger for error correction, and RECAP for failure case optimization. By integrating data preparation, model training, deployment, and cloud inference, the platform addresses computational constraints on robot hardware. Reported applications include logistics operations with throughput exceeding 1200 items/hour and a success rate over 99%, as well as sub-millimeter precision in industrial inspection tasks.