LocoFormer：通过长上下文适应实现通用运动
LocoFormer: Generalist Locomotion via Long-context Adaptation
AI总结：
本文提出通用运动模型 LocoFormer，通过大规模程序化机器人强化学习、激进域随机化和跨回合并列长上下文，实现对未知足式与轮式机器人形态、动力学及故障扰动的测试时适应。
摘要：
现代运动控制器是针对特定本体手动调参的。我们提出 LocoFormer，这是一种通用的全身运动模型，即使在不精确了解其运动学的情况下，也能控制此前未见过的足式和轮式机器人。LocoFormer 能够在测试时适应形态和动力学的变化。我们发现有两个关键选择促成了这种适应。首先，我们在程序化生成的机器人上进行大规模强化学习（RL）训练，并采用激进的域随机化。其次，与以往短上下文长度、缺乏远见的策略不同，我们将上下文延长了数个数量级，使其跨越回合边界。我们将同一个 LocoFormer 部署到各种机器人上，结果表明，即使在重量变化和电机故障等巨大扰动下，它仍能实现鲁棒控制。在极端场景中，我们观察到跨回合的涌现式适应：LocoFormer 从早期回合的跌倒中学习，以改进后续回合的控制策略。我们相信，这一简单而通用的方案未来可用于训练其他机器人技能的基础模型。视频见 generalist-locomotion.github.io。
英文摘要：
Modern locomotion controllers are manually tuned for specific embodiments. We present LocoFormer, a generalist omni-bodied locomotion model that can control previously unseen legged and wheeled robots, even without precise knowledge of their kinematics. LocoFormer is able to adapt to changes in morphology and dynamics at test time. We find that two key choices enable adaptation. First, we train massive scale RL on procedurally generated robots with aggressive domain randomization. Second, in contrast to previous policies that are myopic with short context lengths, we extend context by orders of magnitude to span episode boundaries. We deploy the same LocoFormer to varied robots and show robust control even with large disturbances such as weight change and motor failures. In extreme scenarios, we see emergent adaptation across episodes, LocoFormer learns from falls in early episodes to improve control strategies in later ones. We believe that this simple, yet general recipe can be used to train foundation models for other robotic skills in the future. Videos at generalist-locomotion.github.io.
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
