论文arXiv 2611.231184 天前
ANYmal D世界模型审计:单步验证失效与长时策略部署脱节
本文由 AI 根据公开内容整理,可能有误,请以原文为准:Auditing a Learned Simulator for Legged Robots: The Robotic World Model on ANYmal D(arXiv 2611.23118)
核心内容
针对ANYmal D四足机器人的Robotic World Model (RWM) 及其不确定性估计版本RWM-U,研究团队在无GPU环境下重建并审计了101个训练好的仿真器、200个压缩副本及38个其他架构模型。核心发现是传统单步预测检查严重误判设计选择:虽然单步训练在即时预测上占优,但在368步长时自回归训练中,基于模型自身rollout的训练在所有10个保留测试集中均胜出。此外,剪枝30%权重后单步技能保留90%,但2秒长时技能仅剩39%,证实短时指标对长时任务具有误导性。
在不确定性量化方面,集成分歧度与100步rollout误差Spearman相关系数达+0.90,但因损失函数导致网络方差坍塌,其1-sigma置信区间仅覆盖4.6%结果(校准估计应为68%)。更关键的是仿真精度与策略性能的解耦:18个微调副本在1步和8步预测上均优于原版模型,却有13个生成了失败策略。仅在次级测试中将作者原策略置入各仿真器运行时,才观察到良好的排序一致性,表明当前缺乏无需真实轨迹即可准确评估仿真器策略可用性的可靠指标。
关键看点
- 长时训练优势:基于模型自身rollout的训练在368步 horizon 上全面超越单步训练,证明自回归机制对长时序任务至关重要
- 不确定性校准失效:集成模型虽与长时误差高度相关(+0.90),但因方差坍塌致1-sigma覆盖率仅4.6%,远低于68%的理论值
- 精度-性能脱节:18个预测精度优于基线的微调模型中13个策略部署失败,证实高预测精度不等于有效运动控制
- 模型压缩代价:剪枝30%权重后单步技能保留90%,但2秒长时技能保持率骤降至39%,显示长时动态对参数冗余敏感
工程与产业观察
该审计揭示了具身智能世界模型研发中的“评估陷阱”:单步预测Loss收敛不代表Sim-to-Real可用性。工程实践中应摒弃纯离线单步指标,强制引入长时自回归Rollout测试;同时需警惕不确定性估计在特定Loss下的虚假置信度。对于强化学习训练器选型,建议将“固定策略在仿真器内的表现”作为比“预测误差”更可靠的筛选标准,避免陷入高精度低效能的模型优化死胡同。
英文摘要 (Abstract)
A legged robot can learn to walk inside a learned simulator: a neural network that predicts the robot's next state and is run forward on its own predictions. They are usually trained and checked one step ahead, although a policy uses them for a hundred steps or more. We audit one published system end to end, without a GPU: the dynamics model of the Robotic World Model (RWM) for the ANYmal D quadruped, and the uncertainty estimate of its follow-up, RWM-U. We rebuilt the model to match the released code's outputs, losses and gradients exactly, measured 101 trained simulators, 200 compressed copies and 38 simulators of three other architectures, and trained policies inside simulators that we tested in a hand-built MuJoCo model of the robot. Training on the model's own rollouts beat one-step training at 368 steps on all ten held-out episodes (registered), although one-step training was better one step ahead. One-step checks therefore misjudge design choices: pruning 30% of the weights kept 90% of the model's skill one step ahead and 39% at 2 s. Ensemble disagreement ranked rollout error well (Spearman +0.90 at 100 steps), yet covered 4.6% of outcomes at one sigma, against 68% for a calibrated estimate, as each network's own variance collapses under the implemented loss. Two checks that need no recorded future ranked some pools of simulators well, but neither was shown better than 8-step error in a primary test. At the robot level, accuracy has not predicted walking: all 18 fine-tuned copies we tested predicted better than the released model one and eight steps ahead, yet 13 gave failing policies, and registered tests of whether any check ranks simulators by their policies were not settled. In a secondary test, running the authors' policy inside each simulator ranked them well. We also report nine of our own claims that we withdrew. Scope: one robot, one gait, simulated data, no hardware.
作者:Joyjeet Singh
外部来源
本页事实对应的公开页面,以原文为准。(点击将离开本站)