AS
A.K. Seremak
info
Please Note
<p>This page displays the records of the person named above and is not linked to a unique person identifier. This record may need to be merged to a profile.</p>
1 records found
1
While automated driving for cars has seen significant advancements, the automation of two-wheeled vehicles remains largely underexplored. This gap is primarily due to a lack of appropriate multi-modal datasets and methods that
incorporate bicycle roll, yaw, and forward velocity. Consequently, existing end-to-end (E2E) path planners designed for automobiles cannot be directly applied to the bicycle domain. In this paper we address these limitations by introducing the first comprehensive E2E path planning framework tailored specifically for bicycles. First, we present BikeScenes-planning, a novel multimodal dataset captured using the SenseBike platform, comprising 360-degree RGB images, precise global 2D position, and explicit roll measurements. Second, we demonstrate the necessity of departing from the Average Displacement Error (ADE) metric for bicycle path planner evaluation, and propose a novel, decoupled metric that prioritizes minimizing lateral offset to account for the narrow drivable area in bicycle lanes and the predominant requirement to follow the expert trajectory while allowing for speed deviation. Finally, we establish a baseline for E2E planning that explicitly incorporates roll dynamics into its predictions while using the memory-preserving architectures. Through this work, we pave the way for E2E bicycle path planners, which have the potential to enable fully autonomous two-wheeled vehicles and indirectly improve autonomous driving safety assessments by modeling bicycles as dynamic, reactive agents in simulation evaluations. ...
incorporate bicycle roll, yaw, and forward velocity. Consequently, existing end-to-end (E2E) path planners designed for automobiles cannot be directly applied to the bicycle domain. In this paper we address these limitations by introducing the first comprehensive E2E path planning framework tailored specifically for bicycles. First, we present BikeScenes-planning, a novel multimodal dataset captured using the SenseBike platform, comprising 360-degree RGB images, precise global 2D position, and explicit roll measurements. Second, we demonstrate the necessity of departing from the Average Displacement Error (ADE) metric for bicycle path planner evaluation, and propose a novel, decoupled metric that prioritizes minimizing lateral offset to account for the narrow drivable area in bicycle lanes and the predominant requirement to follow the expert trajectory while allowing for speed deviation. Finally, we establish a baseline for E2E planning that explicitly incorporates roll dynamics into its predictions while using the memory-preserving architectures. Through this work, we pave the way for E2E bicycle path planners, which have the potential to enable fully autonomous two-wheeled vehicles and indirectly improve autonomous driving safety assessments by modeling bicycles as dynamic, reactive agents in simulation evaluations. ...
While automated driving for cars has seen significant advancements, the automation of two-wheeled vehicles remains largely underexplored. This gap is primarily due to a lack of appropriate multi-modal datasets and methods that
incorporate bicycle roll, yaw, and forward velocity. Consequently, existing end-to-end (E2E) path planners designed for automobiles cannot be directly applied to the bicycle domain. In this paper we address these limitations by introducing the first comprehensive E2E path planning framework tailored specifically for bicycles. First, we present BikeScenes-planning, a novel multimodal dataset captured using the SenseBike platform, comprising 360-degree RGB images, precise global 2D position, and explicit roll measurements. Second, we demonstrate the necessity of departing from the Average Displacement Error (ADE) metric for bicycle path planner evaluation, and propose a novel, decoupled metric that prioritizes minimizing lateral offset to account for the narrow drivable area in bicycle lanes and the predominant requirement to follow the expert trajectory while allowing for speed deviation. Finally, we establish a baseline for E2E planning that explicitly incorporates roll dynamics into its predictions while using the memory-preserving architectures. Through this work, we pave the way for E2E bicycle path planners, which have the potential to enable fully autonomous two-wheeled vehicles and indirectly improve autonomous driving safety assessments by modeling bicycles as dynamic, reactive agents in simulation evaluations.
incorporate bicycle roll, yaw, and forward velocity. Consequently, existing end-to-end (E2E) path planners designed for automobiles cannot be directly applied to the bicycle domain. In this paper we address these limitations by introducing the first comprehensive E2E path planning framework tailored specifically for bicycles. First, we present BikeScenes-planning, a novel multimodal dataset captured using the SenseBike platform, comprising 360-degree RGB images, precise global 2D position, and explicit roll measurements. Second, we demonstrate the necessity of departing from the Average Displacement Error (ADE) metric for bicycle path planner evaluation, and propose a novel, decoupled metric that prioritizes minimizing lateral offset to account for the narrow drivable area in bicycle lanes and the predominant requirement to follow the expert trajectory while allowing for speed deviation. Finally, we establish a baseline for E2E planning that explicitly incorporates roll dynamics into its predictions while using the memory-preserving architectures. Through this work, we pave the way for E2E bicycle path planners, which have the potential to enable fully autonomous two-wheeled vehicles and indirectly improve autonomous driving safety assessments by modeling bicycles as dynamic, reactive agents in simulation evaluations.