Tag
AutoEnvScaling treats RL environment design as a terminal task, letting a proposer agent automatically build training environments from solver rollouts to automate the data flywheel. The method raises the valid-environment rate up to 25x at 75% lower cost, improves Qwen terminal-agent scores by up to 12.2 points, and enables a single model to co-evolve as both solver and generator, a step toward recursive self-improvement.
InterMimicGen 提出一个自我进化的运动模仿框架，通过数据飞轮将稀疏的人-物交互动作捕捉数据转化为持续扩展的、可执行的人形机器人全身 loco-manipulation 动作资源，并实现向真实机器人的迁移。
This paper presents Qwen-Planner-Agent, a closed-loop AI-for-AI framework for scalable development of mobile planner agents, integrating data production, training, and deployment to improve performance on real-world tasks and benchmarks.
The tweet highlights Facebook's potential to collect vast amounts of personal AI training data through its 'muse' product, comparing it to Onavo and suggesting it could significantly boost their data flywheel.
This article analyzes three competing approaches in robotics development—foundation models, deployment-first, and full-stack—and suggests that each can lead to significant companies in the large robotics market.
Palantir and NVIDIA partner to deploy Nemotron open models in air-gapped environments for US government agencies, enabling secure, customizable AI with full data control.
FlyRoute is a self-evolving profiling framework that improves LLM-based task routing in multi-agent systems by dynamically updating agent capability descriptions from real traffic, achieving significant accuracy gains over static profiles.