Yuanli Lingji has launched its self-developed embodied AI base model, DM0.5, aiming to address the fragmentation and data challenges in the embodied intelligence sector. The 4B parameter model, trained on 50,000 hours of real-world robot data and 100,000 hours of scene videos, demonstrates significant improvements in zero-shot and few-shot learning capabilities, with a 31% increase in zero-shot navigation success rate and a 45% increase in few-shot success rate compared to its predecessor. The company also introduced the 2.0-Apex general embodied robot platform and the DexDev developer platform, featuring DFOL 2.0, DexOs, and MaaS services to facilitate the practical application and development of embodied AI. AI
IMPACT This release aims to accelerate the practical application of embodied AI by providing a more capable base model and integrated development tools, potentially lowering barriers for industry adoption.
RANK_REASON Frontier-lab model release with system card and platform announcements. [lever_c_demoted from frontier_release: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
