Interventional Behavior Prediction: Avoiding Overly Confident Anticipation in Interactive Prediction

Abstract

Conditional behavior prediction (CBP) builds up the foundation for a coherent interactive prediction and planning framework that can enable more efficient and less conservative maneuvers in interactive scenarios. In CBP task, we train a prediction model approximating the posterior distribution of target agents' future trajectories conditioned on the future trajectory of an assigned ego agent. However, we argue that CBP may provide overly confident anticipation on how the autonomous agent may influence the target agents' behavior. Consequently, it is risky for the planner to query a CBP model. Instead, we should treat the planned trajectory as an intervention and let the model learn the trajectory distribution under intervention. We refer to it as the interventional behavior prediction (IBP) task. Moreover, to properly evaluate an IBP model with offline datasets, we propose a Shapley-value-based metric to testify if the prediction model satisfies the inherent temporal independence of an interventional distribution. We show that the proposed metric can effectively identify a CBP model violating the temporal independence, which plays an important role when establishing IBP benchmarks.

Human-Like Decision Making and Control for Autonomous Driving

“…One prominent issue is that the model is not aware of the AV's future plans and the prediction results are not reactive to the AV's different decisions, forcing the AV to act passively. Some works have recognized this issue and made some efforts to mitigate it [12], [28], [29], [30], [31]. PiP [28] proposes a planninginformed trajectory prediction network that conditions the prediction process on the candidate trajectories of the AV, and [29] formulates such a framework as conditional behavior prediction.…”

“…PiP [28] proposes a planninginformed trajectory prediction network that conditions the prediction process on the candidate trajectories of the AV, and [29] formulates such a framework as conditional behavior prediction. More recently, [31] argues that the AV's future plan should be treated as an intervention rather than an observation in the prediction model. [12] proposes a multiagent policy network as the prediction model that can react to the AV's action.…”

Learning-enabled decision-making for autonomous driving: framework and methodology

“…This can often lead to overly conservative and even dangerous decisions. To overcome this issue, we leverage the conditional motion prediction (CMP) method [87,94,129] in our behavior planning framework, which jointly predicts the future motions of multiple interacting agents in a scene based on the AV's candidate decisions. This provides the evaluation module with more accurate information, enabling improved evaluation of the consequences of different decisions.…”