GAMMA: A General Agent Motion Model for Autonomous Driving

Abstract

Autonomous driving in mixed traffic requires reliable motion prediction of nearby traffic agents such as pedestrians, bicycles, cars, buses, etc.. This prediction problem is extremely challenging because of the diverse dynamics and geometry of traffic agents, complex road conditions, and intensive interactions among the agents. In this paper, we proposed GAMMA, a general agent motion prediction model for autonomous driving, that can predict the motion of heterogeneous traffic agents with different kinematics, geometry, human agents' inner states, etc.. GAMMA formalizes motion prediction as geometric optimization in the velocity space, and integrates physical constraints and human inner states into this unified framework. Our results show that GAMMA outperforms state-of-the-art approaches significantly on diverse real-world datasets.

Closing the Planning-Learning Loop with Application to Autonomous Driving

“…Exo-agents can have two possible motion models. Distracted traffic agents are assumed to track their intended lanes with their current speeds; Attentive traffic agents additionally use GAMMA [18], an optimal collision avoidance model, to simulate the interactions with surrounding agents. Kinematics of all vehicle-like agents are simulated using Bicycle models.…”

“…SUMMIT simulates dense, unregulated urban traffic at worldwide locations supported by the OpenStreetMap. Realistic traffic are automatically generated on urban maps using a traffic motion model [18], the accuracy of which has been validated on various real-world datasets.…”

SUMMIT: A Simulator for Urban Driving in Massive Mixed Traffic

“…Some variants model behavioral types of crowd agents such as patience [26] and attention [27]. A recent model GAMMA [28] can simulate heterogeneous traffic agents with different geometry, kinematics, and behavioral types in a unified, velocity-space framework. The behavior model in SUMMIT extends the framework of GAMMA to encode topological road contexts such as lanes and pedestrian sidewalks to closely represent real-world scenarios.…”

“…SUMMIT uses Context-GAMMA, a context-aware crowd behavior model, to generate sophisticated interactive behaviors of traffic agents. Context-GAMMA extends GAMMA [28] to incorporate road contexts and models them as constraints in velocity space. For completeness, we briefly introduce GAMMA, and present the extensions in Context-GAMMA.…”

“…When computing K A and G τ A , GAMMA also takes into account responsibility and attention of the agent, to generate more human-like motions. We refer readers to [28] for more details of the construction of K A and G τ A . Geometrically, K A is a convex velocity set and G τ A is the intersection of velocity-space half planes.…”