Posts

Table Of Contents Foreword Using NHTSA accidents to improve safety Methodology Step 1: Extracting vehicle shapes from the diagram Step 2: Associate shapes to actual vehicles Step 3: Order waypoints by occurrence Step 4: Generate realistic trajectories Summary (TL;DR;) References Foreword When it comes to safety of autonomous vehicles (AV), we as a society often claim that such a system must be safer as an average human driver. To validate this requirement, the industry pushes metrics such as accident-free travelled distance, expecting to cover the most critical situations an AV would face....

Introduction Evaluating complex systems operating in noisy real world environments is hard. Especially if you are dealing with dynamic behaviour over time, the requirements to the evaluation toolchain grow with every novel insight. In the beginning, you just want to compute just a single number describing algorithm’s performance (like mean squared error - MSE), but over the project’s lifetime your requirements to the evaluation framework grow: at some point you wish to generate static visualizations, create reports over a time range, dig deep at some specific situations or simply render both the algorithm’s output and the corresponding ground-truth (GT) in a video file for human analysis....

TL;DR; This blog post presents the benefits of Frenet-Serret coordinate system for driving trajectory analysis (in Python). This system is defined by a reference path, such as a driving lane and facilitates traffic analysis. As an showcase example, the delay times and distances between vehicles in a roundabout setting are analyzed. The trajectories captured from a bird-view perspective (DJI drone) from the OpenDD dataset are used. In the end of the post, the reader will value the advantages of Frenet-Serret frames compared to a conventional metric approach when it comes to trajectory analysis in a real-world setting....

TLDR; In model based design paradigm one often implements a proof-of-concept solution in a scripting language before porting the logic to a more performant low-level language like C/C++. This article describes the benefits of calling production-ready (C/C++) ports in Python for functional evaluation and analysis. Assessment tools developed previously during the experimentation stage can be reused. Furthermore, the complexity for experimentation with various internal parameters is reduced, which can optimize costs and expensive debugging prior to the final deployment....

Read the article on Medium.

Read the article on Medium.

Read the article on Medium.

Read the article on Medium.