High fidelity behavior prediction of intelligent agents is critical in many applications, which is challenging due to the stochasticity, heterogeneity, and time‐varying nature of agent behaviors. Prediction models that work for one individual may not be applicable to another. Besides, the prediction model trained on the training set may not generalize to the testing set. These challenges motivate the adoption of online adaptation algorithms to update prediction models in real‐time to improve the prediction performance. This article considers online adaptable multitask prediction for both intention and trajectory. The goal of online adaptation is to improve the performance of both intention and trajectory predictions with only the feedback of the observed trajectory. We first introduce a generic ‐step adaptation algorithm of the multitask prediction model that updates the model parameters with the trajectory prediction error in recent steps. Inspired by extended Kalman filter (EKF), a base adaptation algorithm modified EKF with forgetting factor (MEKF) is introduced. In order to improve the performance of MEKF, generalized exponential moving average filtering techniques are adopted. Then this article introduces a dynamic multiepoch update strategy to effectively utilize samples received in real time. With all these extensions, we propose a robust online adaptation algorithm: MEKF with moving average and dynamic multiepoch strategy (MEKF_{MA − ME}). We empirically study the best set of parameters to adapt in the multitask prediction model and demonstrate the effectiveness of the proposed adaptation algorithms to reduce the prediction error.

中文翻译：

---

多任务预测网络的鲁棒非线性自适应算法

智能代理的高保真行为预测在许多应用中至关重要，由于代理行为的随机性，异质性和时变性质，这具有挑战性。适用于一个人的预测模型可能不适用于另一个人。此外，在训练集上训练的预测模型可能不会推广到测试集。这些挑战促使人们采用在线自适应算法来实时更新预测模型以提高预测性能。本文考虑了针对意图和轨迹的在线自适应多任务预测。在线适应的目标是仅通过观察到的轨迹的反馈来提高意图和轨迹预测的性能。我们首先介绍一个通用的多任务预测模型的分步自适应算法，在最近的步骤中使用轨迹预测误差更新模型参数。受扩展卡尔曼滤波器（EKF）的启发，介绍了一种具有遗忘因子（MEKF ）的基本自适应自适应算法EKF 。为了提高MEKF的性能，采用了广义指数移动平均滤波技术。然后，本文介绍了一种动态的多时间段更新策略，可以有效地利用实时接收到的样本。通过所有这些扩展，我们提出了一种强大的在线自适应算法：具有移动平均和动态多周期策略的MEKF（MEKF _MA-ME）。我们根据经验研究了适合多任务预测模型的最佳参数集，并证明了所提出的自适应算法减少预测误差的有效性。