让轨迹预测模型能听懂指令，还能自适应场景变化。
RetroMotion: Retrocausal Motion Forecasting Models are Instructable
- 用变换器分步建模：先算个体轨迹，再结合交互关系生成联合分布。
- 在Waymo、Argoverse 2等数据集上表现优异，预测精度领先。
- 训练时隐含学会理解指令，可动态调整行为响应场景需求。
道路使用者（即智能体）的运动预测复杂度随智能体数量、场景约束和交互关系呈指数增长。为此，我们把多智能体预测分解为（1）所有建模智能体的边缘分布，和（2）相互作用智能体的联合分布。采用变换器模型，通过重编码边缘分布并引入成对建模生成联合分布，实现从后期边缘轨迹向早期联合轨迹的信息回溯（retrocausal flow）。每个时间步使用压缩幂指数分布建模位置不确定性。实验表明，该方法在Waymo交互预测挑战赛中表现强劲，并在Argoverse 2和V2X-Seq数据集上具有良好泛化能力。此外，模型具备指令接口：标准运动预测训练过程隐式赋予模型理解与适配指令的能力，使其可根据上下文动态调整预测行为。
原文摘要 · Abstract (English)
Motion forecasts of road users (i.e., agents) vary in complexity depending on the number of agents, scene constraints, and interactions. In particular, the output space of joint trajectory distributions grows exponentially with the number of agents. Therefore, we decompose multi-agent motion forecasts into (1) marginal distributions for all modeled agents and (2) joint distributions for interacting agents. Using a transformer model, we generate joint distributions by re-encoding marginal distributions followed by pairwise modeling. This incorporates a retrocausal flow of information from later points in marginal trajectories to earlier points in joint trajectories. For each time step, we model the positional uncertainty using compressed exponential power distributions. Notably, our method achieves strong results in the Waymo Interaction Prediction Challenge and generalizes well to the Argoverse 2 and V2X-Seq datasets. Additionally, our method provides an interface for issuing instructions. We show that standard motion forecasting training implicitly enables the model to follow instructions and adapt them to the scene context. GitHub repository: https://github.com/kit-mrt/future-motion
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。