Decentralized Vehicle Coordination: The Berkeley DeepDrive Drone Dataset and Consensus-Based Models

TL;DR

The paper addresses collision avoidance on understructured roads where right-of-way is informal by introducing the Berkeley DeepDrive Drone Dataset (B3D) and a consensus-based, least-action framework for decentralized coordination. The approach replaces predict-then-plan with negotiate-then-commit, using local sensing and a cost-based optimization to determine priority orders among $N$ agents with intersecting paths. Key contributions include publicly releasing 20 aerial videos, 16002 annotated images, and a trajectory-estimation development kit, plus two-agent validation and integrative $N$-agent simulations that demonstrate the model's ability to reproduce and manage decentralized conflict resolution. The work advances understanding of social driving etiquette in understructured environments and provides data and tools to develop decentralized planning methods with potential applicability to other mobile robots and autonomous systems.

Abstract

A significant portion of roads, particularly in densely populated developing countries, lacks explicitly defined right-of-way rules. These understructured roads pose substantial challenges for autonomous vehicle motion planning, where efficient and safe navigation relies on understanding decentralized human coordination for collision avoidance. This coordination, often termed "social driving etiquette," remains underexplored due to limited open-source empirical data and suitable modeling frameworks. In this paper, we present a novel dataset and modeling framework designed to study motion planning in these understructured environments. The dataset includes 20 aerial videos of representative scenarios, an image dataset for training vehicle detection models, and a development kit for vehicle trajectory estimation. We demonstrate that a consensus-based modeling approach can effectively explain the emergence of priority orders observed in our dataset, and is therefore a viable framework for decentralized collision avoidance planning.

Figures (5)

• Figure 1: Tailgating collisions in and . The collided vehicles are circled in black. The first accident involves at least two vehicles, while the second incident involves four vehicles.
• Figure 2: Timeline of the tailgating accidents. Green indicates regular traffic. Red indicates congestion caused by a collision. Light red indicates the induced congestion starts dissipating.
• Figure 3: Timeline of the stop-and-go waves. Green indicates regular traffic. Red indicates congestion caused by a strong stop-and-go wave. Light red indicates congestion caused by a weak stop-and-go wave.
• Figure 4: The $N$-agent simulation scenario. Red circles: trip origins. Black circles: destinations. Numbered green rectangles: vehicles at the intersection.
• Figure 5: Results of the $N$-agent simulation. Red points: southbound vehicles. Blue points: westbound vehicles.

Theorems & Definitions (4)

• Remark 1
• Remark 2
• Remark 3
• Remark 4