The Oxford RobotCar Dataset is a large-scale autonomous driving dataset collected by repeatedly driving a route through central Oxford between May 2014 and December 2015. It comprises over 1,000 km of driving across more than 1,000 traversals, capturing nearly 20 million images from six vehicle-mounted cameras, along with data from LIDAR, GPS, and INS systems. These traversals span diverse environmental conditions—including heavy rain, nighttime, direct sunlight, snow, and structural changes like roadworks—making the dataset a valuable resource for long-term localization and mapping in dynamic urban environments.
Use case:
Autonomous Vehicle Localization and Mapping in Dynamic Urban Environments
| Example Observation | – Image frontale (RGB) ou embedding CNN (256 d). – Vitesse véhicule (m/s) et accélération long. (m/s²). – Angle volant courant (rad). – IMU : lacet, tangage, roulis. – GPS/INS position (lat, lon) et heading. – Horodatage + conditions météo (pluie, nuit / jour) dérivables via sessions. |
| Example Action | – δₜ : commande de braquage (continu : −0.6 rad ↔ +0.6 rad). – αₜ : consigne d’accélération (continu : −4 m/s² freinage ↔ +3 m/s²). |
| Example Reward | rₜ = − w₁ · |déviation latérale| − w₂ · |surdépassement vitesse| − w₃ · |accélération longitudinale| − w₄ · collisions. – Les collisions peuvent être approchées par des pics de freinage brusque (> −4 m/s²) ou des distances lidar < 0.5 m ; pénalité −1. – Pondérations typiques : w₁=0.2, w₂=0.1, w₃=0.05, w₄=1.0. |