Papers
Topics
Authors
Recent
Search
2000 character limit reached

Online Vehicle Trajectory Prediction using Policy Anticipation Network and Optimization-based Context Reasoning

Published 3 Mar 2019 in cs.RO and cs.LG | (1903.00847v1)

Abstract: In this paper, we present an online two-level vehicle trajectory prediction framework for urban autonomous driving where there are complex contextual factors, such as lane geometries, road constructions, traffic regulations and moving agents. Our method combines high-level policy anticipation with low-level context reasoning. We leverage a long short-term memory (LSTM) network to anticipate the vehicle's driving policy (e.g., forward, yield, turn left, turn right, etc.) using its sequential history observations. The policy is then used to guide a low-level optimization-based context reasoning process. We show that it is essential to incorporate the prior policy anticipation due to the multimodal nature of the future trajectory. Moreover, contrary to existing regression-based trajectory prediction methods, our optimization-based reasoning process can cope with complex contextual factors. The final output of the two-level reasoning process is a continuous trajectory that automatically adapts to different traffic configurations and accurately predicts future vehicle motions. The performance of the proposed framework is analyzed and validated in an emerging autonomous driving simulation platform (CARLA).

Citations (49)

Summary

Online Vehicle Trajectory Prediction Using Policy Anticipation Network and Optimization-Based Context Reasoning

This paper presents an innovative approach to vehicle trajectory prediction in urban environments, addressing the complexities associated with dynamic contextual factors such as road geometry, traffic regulations, and interactions with other moving agents. The authors propose a two-level prediction framework combining high-level policy anticipation with low-level optimization-based context reasoning. This solution critically addresses the multimodal nature of future trajectories and the intricate dependencies on environmental conditions.

Framework Overview

The proposed framework consists of a high-level policy anticipation network combined with a low-level optimization-based reasoning process. The policy anticipation network utilizes a Long Short-Term Memory (LSTM) architecture to predict high-level driving policies such as moving forward, yielding, turning, and lane changing based on sequential historical observations. This high-level predicted policy informs the subsequent optimization process, which generates trajectory predictions by balancing various contextual costs within an optimization framework.

Key Contributions and Methodology

Policy Anticipation Network

The use of an LSTM-based policy anticipation network enables the system to predict likely future policies based on a sliding window of historical vehicle states. This approach allows the anticipation of complex maneuvers in scenarios such as intersections with differing directional options. The network outputs a probability distribution across policy labels to handle uncertainty and variations in vehicle behaviors.

Optimization-Based Context Reasoning

Central to the framework's low-level reasoning is a multi-layer cost map structure encoding diverse driving context factors, including lane geometry, static and dynamic obstacles, traffic lights, and speed limits. Non-holonomic constraints are considered to ensure compliance with realistic vehicular dynamics. The optimization seeks a trajectory minimizing a weighted sum of these costs, effectively integrating multimodal environmental influences. This structured problem formulation allows for adaptability across various urban driving conditions.

Validation and Results

Testing was conducted using the autonomous driving simulator CARLA, providing realistic urban scenario data. The framework demonstrated effectiveness across various traffic configurations: curved roads, intersections with construction, and interactions with pedestrians. Quantitative evaluations revealed significant accuracy improvements over baseline models that lacked context-awareness and multimodal integration. The proposed method consistently achieved lower prediction errors, confirming the importance of a hierarchical reasoning strategy encompassing intention inference and contextual adaptability.

Implications and Future Directions

The results indicate strong potential for deploying this framework in real-world autonomous vehicle systems, particularly in complex urban environments. By effectively predicting vehicle trajectories that account for high-level intentions and contextual factors, this method enhances safety and efficiency in autonomous navigation. Future work may focus on integrating inverse reinforcement learning (IRL) for automated tuning of cost weights, further refining the framework’s capability to model human driver behavior. Additionally, exploration into incorporating real-time interaction modeling between agents holds promise for advancing trajectory prediction methodologies.

The study provides a versatile and robust approach to trajectory prediction, recognizing both theoretical implications for modeling decision-making processes and practical benefits for autonomous driving technologies.

Paper to Video (Beta)

No one has generated a video about this paper yet.

Whiteboard

No one has generated a whiteboard explanation for this paper yet.

Open Problems

We haven't generated a list of open problems mentioned in this paper yet.

Continue Learning

We haven't generated follow-up questions for this paper yet.

Authors (2)

Collections

Sign up for free to add this paper to one or more collections.