Act Natural! Projecting Autonomous System Trajectories Into Naturalistic Behavior Sets

TL;DR

The paper tackles enabling autonomous systems to operate in a human-friendly manner by projecting trajectories onto a naturalistic behavior set defined by time-indexed convex hulls $N_t$. It frames this as a convex optimization with linear dynamics $\mathbf{x}_{t+1} = A\mathbf{x}_t + B\mathbf{u}_t$ and linear constraints $G_t y_t \le h_t$, minimizing the distance to the original path $\xi_a$ via $d(\cdot,\cdot)$. The naturalistic sets $N_t$ are constructed from real driving data using planar double-integrator dynamics and convex hulls computed with Quickhull, and validated through curved-road and busy-intersection case studies, showing the projection yields human-like, dynamically feasible trajectories without explicit intent modeling. The work highlights advantages in data efficiency and simplicity, while acknowledging limitations from non-convexity, interaction, and potential need for unions or discrete decision-making, outlining directions for improving expressiveness and safety in autonomous planning.

Abstract

Autonomous agents operating around human actors must consider how their behaviors might affect those humans, even when not directly interacting with them. To this end, it is often beneficial to be predictable and appear naturalistic. Existing methods to address this problem use human actor intent modeling or imitation learning techniques, but these approaches rarely capture all possible motivations for human behavior or require significant amounts of data. In contrast, we propose a technique for modeling naturalistic behavior as a set of convex hulls computed over a relatively small dataset of human behavior. Given this set, we design an optimization-based filter which projects arbitrary trajectories into it to make them more naturalistic for autonomous agents to execute while also satisfying dynamics constraints. We demonstrate our methods on real-world human driving data from the inD intersection dataset (Bock et al., 2020).

Figures (1)

• Figure 2: We compute the naturalistic behavior set over two-dimensional position using all trajectories of moving vehicles that begin in the green square and end in the red. We generate the naturalistic behavior set based on these trajectories and plot it. Parked cars along the driving lane are boxed in light green. We additionally circle (in light blue) the area of the lane in which vehicles tend to yield to crossing road users.