Li Zhang (lizhangfd@fudan.edu.cn) is the corresponding author. Previous methods, as depicted in (a), use only one mode query for each trajectory.

By pairing historical and future trajectories and applying contrastive learning on the encoded feature space, we enforce same-space consistency constraints.

Neural Information Processing Systems http://nips.cc/

In other words, a person's subsequent trajectory has likely been traveled by others. Based on this hypothesis, we propose to forecast a person's future trajectory by learning from the implicit scene regularities. We call the regularities, inherently derived from the past dynamics of the people and the environment in the scene, scene history.

This work has the potential for wide-ranging applications in human-in-the-loop (e.g. We set the radius of agents to 0.3, the radius of The dataset will be made public. The only difference of our model's architecture to theirs is that we use agent-centric representations Then, we construct an edge from the agent that corresponds to the row to the "column agent" then compare this with the ground truth graph. The smaller the circle, the further it is into the future.

We introduce how to compare baselines with our proposed method in Table 1 and 2. Since all MOE [ Therefore, our aim is to calculate the third term. We show the analytical solution for the KL divergence of two multivariate Gaussian distributions. We will further optimize our method according to these failure cases in our future work. The ground-truth future trajectories are shown in orange color. In addition, we also draw the predicted unobserved trajectories (blue) by our BCDiff and ground-truth (cyan).

Therefore, it is necessary to perform trajectory prediction based on instantaneous observations, which has rarely been studied before.

Modeling the behavior of intelligent agents is an essential subject for autonomous systems. Safe operations of autonomous agents require accurate prediction of other agents' future motions.