SafeVRU: A Research Platform for the Interaction of Self-Driving Vehicles with Vulnerable Road Users

Abstract

This paper presents our research platform SafeVRU for the interaction of self-driving vehicles with Vulnerable Road Users (VRUs, i.e., pedestrians and cyclists). The paper details the design (implemented with a modular structure within ROS) of the full stack of vehicle localization, environment perception, motion planning, and control, with emphasis on the environment perception and planning modules. The environment perception detects the VRUs using a stereo camera and predicts their paths with Dynamic Bayesian Networks (DBNs), which can account for switching dynamics. The motion planner is based on model predictive contouring control (MPCC) and takes into account vehicle dynamics, control objectives (e.g., desired speed), and perceived environment (i.e., the predicted VRU paths with behavioral uncertainties) over a certain time horizon. We present simulation and real-world results to illustrate the ability of our vehicle to plan and execute collision-free trajectories in the presence of VRUs. I. INTRODUCTION Every year between 20 and 50 million people are involved in road accidents, mostly caused by human errors [1]. According to [1], approximately 1.3 million people lost their life in these accidents. Half of the victims are vulnerable road users (VRUs), such as pedestrians and cyclists. Self-driving vehicles can help reduce these fatalities [2]. Active safety features, such as autonomous emergency braking (AEB), are increasingly found on-board vehicles on the market to improve VRUs' safety (see [3] for a recent overview). In addition, some vehicles already automate steering functionality (e.g., [4], [5]), but still require the driver to initiate the maneuver. Major challenges must be addressed to ensure safety and performance while driving in complex urban environments [6], where VRUs are present. The self-driving vehicle should be aware of the presence of the VRUs and be able to infer their intentions to plan its path accordingly to avoid collisions. In this respect, motion planning methods are required to provide safe (collision-free) and systemcompliant performance in complex environments with static and moving obstacles (refer to [7], [8] for an overview). In real-world applications, the information on the pose (i.e., position and orientation) of other traffic participants comes from a perception module. The perception module should provide to the planner information not only concerning the current position of the other road users, but also † The authors equally contributed to the paper.

CNN Based Road User Detection Using the 3D Radar Cube

“…Our real-world dataset contains ∼ 1 hour of driving in urban environment with our demonstrator vehicle [26]. We recorded both the target-level and low-level output of our radar, a Continental 400 series mounted behind the front bumper.…”

Model Predictive Contouring Control for Collision Avoidance in Unstructured Dynamic Environments

“…The accompanying video shows the results where the autonomous vehicle successfully avoids the moving obstacles, while staying within the road limits. We refer the reader to [32] for more details and results.…”

Inter-Urban Analysis of Pedestrian and Drivers through a Vehicular Network Based on Hybrid Communications Embedded in a Portable Car System and Advanced Image Processing Technologies

“…We can say that during all this time the problem of detection is concentrated in the form of a convolutional neural network, concatenating the output data with the inputs from the original image by parsing the location of the characteristics. Therefore, we consider that the detection frames of those faces depend on the received coordinates and outline the last prediction frame for all frames in the created grid, so the coordination values of the anchor box are randomly reset and the network is fully examined to obtain a punctual and concrete analysis between the detection distance and the predictability area, as opposed to the standard defined by Equation (6), where IOU are characterized by the ratio between the intersections of unique feature sets and the real area.…”

“…The literature includes studies and analysis in this field since the 1970s, especially the Oriented Gradient Histogram (HOG), this being an efficient way to extract important features from an image, then obtaining a model that can be classified and later recognized as a series of objects, and here we can refer to some of the works of Badler and Smoilar [4]. Later approaches and future directions appeared in Badler's article [5] and that of Gavrila [6,7]. In addition, collaborations in the matter of analyzes and practical demonstrations by Mikolajczyk et al [8] later adapted other elements, in terms of detection, from Dalal and Triggs [9,10].…”