Skip to content

odometry_estimation.md

Latest commit

 

History

History
36 lines (28 loc) · 2.24 KB

odometry_estimation.md

File metadata and controls

36 lines (28 loc) · 2.24 KB
layout default
title Odometry Estimation

Multimodal Off-Road Odometry for Robust Localization

Overview

This challenge aims to address one of the core difficulties in off-road autonomy: estimating vehicle odometry in environments where traditional GPS-based methods may be unreliable or degraded. Participants must use multimodal sensor data to estimate the vehicle’s trajectory while overcoming challenges such as sensor noise, GPS dropouts, and rough, unpredictable terrain.

Task Details

Participants are tasked with developing algorithms that fuse multiple sensor modalities to accurately estimate vehicle odometry in off-road environments. The dataset includes synchronized data from various sensors:

  • RGB Cameras: For visual odometry and terrain recognition.
  • LiDAR: For detecting the vehicle’s surroundings and mapping terrain structure.
  • RADAR: For understanding the environment in adverse conditions (e.g., fog, dust).
  • Thermal: To capture the heat signatures of the environment.
  • IMU and GPS: While GPS data will be included, some portions of the dataset will simulate degraded GPS signals, forcing participants to rely on other sensor data.

Participants can explore techniques like:

  • Sensor Fusion Algorithms: Combining data from multiple sources to improve odometry accuracy, especially in GPS-denied areas.
  • Deep Learning for Odometry: Learning-based methods to estimate trajectory from sensor data.
  • SLAM (Simultaneous Localization and Mapping): Developing or adapting SLAM algorithms to off-road environments where terrain features are less structured.

Dataset

Participants will have access to:

  • Multimodal sensor data collected from off-road environments.
  • Synchronized data streams from all sensors to enable efficient fusion and localization.

Submission:

Participants will submit estimated odometry along with a brief report describing their approach and how they handled off-road challenges and sensor fusion.

Evaluation Metrics

The submissions will be evaluated based on:

  • Root Mean Square Error (RMSE): Measures the accuracy of the estimated trajectory relative to the ground truth.
  • Absolute Trajectory Error (ATE): Assesses the global consistency of the estimated trajectory.