NVIDIA 把 GPT 的套路用在运动控制上，预训练一个模型就能微调干不同活，物理模拟里跑得又自然又实时。
NVIDIA 在 SIGGRAPH 上发表 Generative Pretrained Controllers (GPC)，将运动技能编码为离散 token。GPC 使用 transformer 进行下一个 token 预测，类似 GPT 的预训练范式。它在 600+小时运动数据上训练，可实时运行在物理模拟中。同一个预训练控制器通过微调即可适配新任务，生成自然且物理合理的交互动作。
Most motion papers tailor one controller to one specific task. This year at SIGGRAPH, our research t...
Most motion papers tailor one controller to one specific task. This year at SIGGRAPH, our research team asks: can motor control itself be pretrained and reused? Generative Pretrained Controllers, or GPC, turn motor skills into a vocabulary of discrete tokens and train a transformer-based generative controller through next-token prediction. Just like GPT, the same pretrained controller can then be fine-tuned to solve new tasks. Trained on 600+ hours of motion, GPC runs in real-time inside a physics simulation, producing natural and physically grounded behaviors for interactive control. Your browser does not support the video tag. 🔗 View on Twitter 💬 5 🔄 8 ❤️ 54 👀 4108 📊 13 ⚡