Social-Implicit: Rethinking Trajectory Prediction Evaluation and The Effectiveness of Implicit Maximum Likelihood Estimation

Understanding Pedestrian Trajectory Prediction from the Planning Perspective

“…In our experiments, two publicly available datasets named the ETH [23] and UCY [24] datasets are utilized. The ETH dataset consists of two distinct scenarios, named the Hotel and ETH scenarios, while the UCY dataset contains three scenarios named ZARA1, ZARA2, and UNIV.…”

A Review of Tracking and Trajectory Prediction Methods for Autonomous Driving

“…Since then, limited progress in enhancing the prediction accuracy has prompted researchers to explore other crucial factors of trajectory prediction methods. As a result, Social-Implicit [1] was introduced in 2022, leveraging feed-forward neural networks in place of recurrent ones. This shift yielded enhancements in computational efficiency without compromising on the model's accuracy, thereby contributing to the progression in this field.…”

“…A VOIDING accidents involving vulnerable road users (VRU), such as pedestrians, is one of the paramount objectives and ongoing challenges for autonomous vehicles. Consequently, the prediction of pedestrian behavior remains an active area of research, as evidenced by recently published methods [1], [2], [3] and review papers [4], [5]. One of the many reasons for the ongoing research effort are conditions encountered in real-world scenarios, which are characterized by imperfect observations and the stochastic nature of human behavior [6], [5].…”

“…In addition, interactions can be explicitly considered. Most commonly, only the last eight timesteps are provided (equivalent to 3.2 s) and based on that, 12 timesteps are predicted [8], [9], [10], [1], [11]. Initially modeled through simple rule-based methods [12], various architectures based on neural networks Nico Uhlemann, Felix Fent and Markus Lienkamp are with the Technical University of Munich, Germany; School of Engineering & Design, Institute of Automotive Technology and Munich Institute of Robotics and Machine Intelligence (MIRMI) as well as hybrid approaches [2], [3] have emerged in recent years, trying to improve the overall accuracy of the predicted paths on widely-used benchmarks like ETH/UCY [13], [14] and SSD [15].…”

“…In this paper, we want to build upon the work outlined in previous publications [1], [16], [6], evaluating the state of the art to derive the reliability as well as the applicability of these models when confronted with situations encountered on the road. Determining the accuracy, feature requirements, and computational efficiency, the contributions of this work are therefore threefold:…”