[Paper Review] INTERACTION Dataset: An INTERnational, Adversarial and Cooperative moTION Dataset in Interactive Driving Scenarios with Semantic Maps

Wei Zhan, Liting Sun+9 more|arXiv (Cornell University)|Sep 30, 2019

Autonomous Vehicle Technology and Safety43 references354 citations

TL;DR

The paper introduces the INTERACTION dataset, a drone- and camera-recorded, international, highly interactive driving motion dataset with semantic HD maps, designed for motion prediction, planning, imitation learning, and behavior analysis.

ABSTRACT

Behavior-related research areas such as motion prediction/planning, representation/imitation learning, behavior modeling/generation, and algorithm testing, require support from high-quality motion datasets containing interactive driving scenarios with different driving cultures. In this paper, we present an INTERnational, Adversarial and Cooperative moTION dataset (INTERACTION dataset) in interactive driving scenarios with semantic maps. Five features of the dataset are highlighted. 1) The interactive driving scenarios are diverse, including urban/highway/ramp merging and lane changes, roundabouts with yield/stop signs, signalized intersections, intersections with one/two/all-way stops, etc. 2) Motion data from different countries and different continents are collected so that driving preferences and styles in different cultures are naturally included. 3) The driving behavior is highly interactive and complex with adversarial and cooperative motions of various traffic participants. Highly complex behavior such as negotiations, aggressive/irrational decisions and traffic rule violations are densely contained in the dataset, while regular behavior can also be found from cautious car-following, stop, left/right/U-turn to rational lane-change and cycling and pedestrian crossing, etc. 4) The levels of criticality span wide, from regular safe operations to dangerous, near-collision maneuvers. Real collision, although relatively slight, is also included. 5) Maps with complete semantic information are provided with physical layers, reference lines, lanelet connections and traffic rules. The data is recorded from drones and traffic cameras. Statistics of the dataset in terms of number of entities and interaction density are also provided, along with some utilization examples in a variety of behavior-related research areas. The dataset can be downloaded via https://interaction-dataset.com.

Motivation & Objective

• Provide a large-scale, internationally sourced dataset of interactive driving scenarios.
• Capture diverse, complex, and critical interactions including adversarial and cooperative behaviors.
• Include complete semantic high-definition maps (lanelets, rules, references) and full interaction entities.
• Enable studies in motion prediction, imitation learning, decision-making, planning and social-behavior generation.

Proposed method

• Collect interactive driving data from drones and traffic cameras across multiple countries and continents.
• Annotate trajectories with accurate bounding boxes and ground-plane trajectories using stabilization, detection (Faster R-CNN), data association, tracking (Kalman) and smoothing (RTS).
• Construct centimeter-accurate high-definition lanelet2 maps with physical and semantic layers (lanelets, rules, right-of-way).
• Provide diverse scenarios including roundabouts, unsignalized and signalized intersections, merges and lane changes.
• Evaluate interaction density using metrics like minimum time-to-conflict-point differences and waiting periods to identify interaction pairs.

Experimental results

Research questions

• RQ1 How does driving behavior vary across international contexts in highly interactive driving scenarios?
• RQ2 Can high-density interactive trajectories with semantic maps improve prediction, planning and imitation learning models?
• RQ3 What are the characteristics and distribution of critical/interacting events (near-collisions, aggressive maneuvers) in diverse scenarios?
• RQ4 How does the availability of complete interaction entities and maps affect modeling and planning performance?

Key findings

• The dataset includes diverse scenarios such as roundabouts, ramps, unsignalized and signalized intersections across multiple continents.
• It captures highly interactive and complex behaviors, including adversarial and cooperative motions, with instances of near-collision and slight collision events.
• HD maps with complete semantic information are provided, enabling semantically informed prediction and planning.
• Interaction density metrics reveal higher interaction intensity in INTERACTION than in prior datasets like highD and NGSIM, especially for short TTCP differences (<1 s).
• The data supports usage in motion prediction, imitation learning, decision-making and planning validation, as well as interaction extraction and social behavior generation.