Act Natural! Extending Naturalistic Projection to Multimodal Behavior Scenarios
Hamzah I. Khan, David Fridovich-Keil
TL;DR
This work addresses the challenge of achieving naturalistic, predictable behavior for autonomous agents in human environments by extending unimodal naturalistic projection to multimodal scenarios. It introduces a data-driven representation where time-indexed naturalistic behavior at each moment is modeled as a union of convex hulls, learned from observed trajectories through clustering and hull construction, and couples this with a mixed-integer optimization to project arbitrary trajectories into the learned set while preserving dynamics. Key contributions include (i) a multimodal naturalistic set identification framework, (ii) an online trajectory projection method that enforces naturalistic constraints via binary selection of convex hulls, and (iii) comprehensive experiments on inD and rounD datasets across curved roads, intersections, and roundabouts, with analysis of runtime-accuracy tradeoffs. The approach provides a practical, interpretable, data-efficient alternative to purely data-hungry imitation-learning methods, enabling real-time planning that respects human-like behavior in complex driving scenarios.
Abstract
Autonomous agents operating in public spaces must consider how their behaviors might affect the humans around them, even when not directly interacting with them. To this end, it is often beneficial to be predictable and appear naturalistic. Existing methods for this purpose use human actor intent modeling or imitation learning techniques, but these approaches rarely capture all possible motivations for human behavior and/or require significant amounts of data. Our work extends a technique for modeling unimodal naturalistic behaviors with an explicit convex set representation, to account for multimodal behavior by using multiple convex sets. This more flexible representation provides a higher degree of fidelity in data-driven modeling of naturalistic behavior that arises in real-world scenarios in which human behavior is, in some sense, discrete, e.g. whether or not to yield at a roundabout. Equipped with this new set representation, we develop an optimization-based filter to project arbitrary trajectories into the set so that they appear naturalistic to humans in the scene, while also satisfying vehicle dynamics, actuator limits, etc. We demonstrate our methods on real-world human driving data from the inD (intersection) and rounD (roundabout) datasets.