[Paper Review] Motion Prediction on Self-driving Cars: A Review

Shahrokh Paravarzar, Belqes Mohammad|arXiv (Cornell University)|Nov 6, 2020

Autonomous Vehicle Technology and Safety29 references4 citations

TL;DR

This review synthesizes state-of-the-art approaches to motion prediction in autonomous vehicles, evaluating classical, deep learning, and reinforcement learning methods. It identifies deep reinforcement learning as the most promising approach due to its ability to model complex, multimodal interactions under dynamic traffic conditions.

ABSTRACT

The autonomous vehicle motion prediction literature is reviewed. Motion prediction is the most challenging task in autonomous vehicles and self-drive cars. These challenges have been discussed. Later on, the state-of-theart has reviewed based on the most recent literature and the current challenges are discussed. The state-of-the-art consists of classical and physical methods, deep learning networks, and reinforcement learning. prons and cons of the methods and gap of the research presented in this review. Finally, the literature surrounding object tracking and motion will be presented. As a result, deep reinforcement learning is the best candidate to tackle self-driving cars.

Motivation & Objective

• To analyze the current state of motion prediction in autonomous vehicles, focusing on challenges in modeling dynamic, interactive traffic behavior.
• To evaluate the strengths and limitations of classical, deep learning, and reinforcement learning-based motion prediction methods.
• To identify research gaps in accuracy, safety, and generalization required for real-world deployment of self-driving systems.
• To examine the role of semantic maps, sensor fusion, and environmental constraints in shaping prediction performance.
• To assess the viability of reinforcement learning as a scalable solution for handling multimodal and interdependent vehicle behaviors.

Proposed method

• Systematic review of literature from 2011 to 2020 on motion prediction in autonomous vehicles, categorized by methodological approach.
• Classification of motion prediction techniques into three main categories: classical/physical models, deep learning (CNN-RNN hybrids), and reinforcement learning.
• Evaluation of methods based on their ability to handle multimodal behavior, environmental dependencies, and real-time constraints.
• Analysis of sensor data integration (LiDAR, radar, cameras) and semantic mapping for context-aware prediction.
• Examination of motion modeling using probabilistic frameworks and trajectory forecasting under uncertainty.
• Incorporation of traffic rules, road geometry, and dynamic environmental changes as constraints in predictive models.

Experimental results

Research questions

• RQ1 How do classical and physical motion prediction models compare in handling complex, real-world traffic interactions?
• RQ2 To what extent do deep learning models (e.g., CNN-RNN) improve prediction accuracy over traditional methods in multimodal scenarios?
• RQ3 Can reinforcement learning models effectively capture long-term dependencies and inter-vehicle dependencies in autonomous driving?
• RQ4 What are the key limitations in current motion prediction models regarding safety, robustness, and generalization across diverse driving environments?
• RQ5 How do environmental factors such as weather, road geometry, and traffic rules influence the performance of motion prediction systems?

Key findings

• Deep learning models, particularly those combining CNNs and RNNs, demonstrate improved performance in modeling spatial-temporal dependencies in vehicle trajectories.
• Reinforcement learning methods show superior potential for handling complex, multimodal behaviors and inter-vehicle interactions compared to classical and deep learning baselines.
• Despite progress, existing models still fall short in achieving the required accuracy and safety standards for real-world deployment in high-risk traffic scenarios.
• The integration of semantic maps and sensor fusion significantly enhances context awareness and prediction reliability in dynamic environments.
• Multimodal behavior—such as lane changes, stops, and sudden maneuvers—remains a major challenge due to the high uncertainty and interdependence among vehicles.
• Current models often fail to generalize across diverse driving conditions, especially under adverse weather or unexpected road changes.