Training AI for Space:

on YouTube ↗21m

" summary

Claims a production pipeline that runs multi-agent, multi-mission RL training on a serialized, configurable RL Gym that swaps simulation engines via a simple API, uses Ray + RLlib on AnyScale for distributed runs, and packs each agent with its simulation so they run as many as 600 simultaneous simulations (one per core) to raise frames-per-second. Agents output a six-dimensional orbit vector of intended state; action translators convert intents into optimal-control commands. Demos include Kerbal Space Program showing an imaging task; weights are not shared publicly.

AI-written summary · about

" chapters (7)

1. 0:00Introduction to talk
2. 1:04Why reinforcement learning for space
3. 2:51Challenges in RL for space flight
4. 5:00Scaling RL with Ray and RLlib
5. 6:26Gym architecture for modular training
6. 8:33Demo of RL agent in Kerbal Space
7. 11:49Q&A session

" related talks

• 15mRollArt: Disaggregated Multi-Task Agentic RL Training at ScaleWei Wang, Tianyuan Wu, Lunxi Cao, Siran Yang, Jiamang Wang, Lin Qu, Wei Gao, Yuheng Zhao, Dakai An, Shaopan Xiong, Ju Huang, Weixun Wang, Wenbo Su, Bo Zheng, Dilxat Muhtar, Zichen Liu, Haizhou Zhao, Yongbin Li · 20th USENIX Symposium on Operating Systems Design and Implementation (OSDI '26)
• 19mDynaRL: Flexible and Dynamic Scheduling of Large-Scale Reinforcement Learning TrainingYuanqing Wang, Yuchen Zhang, Yu Wang, Junhao Hu, Chao Yu, Hao Lin, Zhi Yang, Zhen Guo, Si Xu, Quanlu Zhang, Chunyang Zhu, Zixiao Huang, Bo Dai, Xu Fu, Boxun Li, Guohao Dai · 20th USENIX Symposium on Operating Systems Design and Implementation (OSDI '26)
• 28mScaling Multimodal Data and Reinforcement Learning with RayJaikumar Ganesh · AI By the Bay 2025
• 14mConnecting Reinforcement Learning to Simulation SoftwareMax Pumperla · Ray Summit 2020
• 01h 40mOrchestrating a Role-Playing Game Universe with Multi-Container AI AgentsPhilippe Charrière, Guillaume Lours · DevFest Toulouse 2025
• 17mRLinf: Flexible and Efficient Large-Scale Reinforcement Learning...Yuanqing Wang, Zhihao Liu, Junhao Hu, Yongji Wu, Chao Yu, Yuqing Xie, Kang Chen, Hao Lin, Zhen Guo, Si Xu, Hongzhi Zang, Quanlu Zhang, Chunyang Zhu, Zixiao Huang, Mingjie Wei, Ke Yang, Bo Dai, Zhexuan Xu, Jiakun Du, Xiangyuan Wang, Xu Fu, Letong Shi · 20th USENIX Symposium on Operating Systems Design and Implementation (OSDI '26)
• 19mA 24x Speedup for Reinforcement Learning with RLlib + RayRaoul Khouri · Ray Summit 2021
• 26mBeyond Sandboxes: Architecting Durable Runtimes for AI AgentsViren Baraiya · QCon AI Boston 2026