Bot-VIO

BotVIO: A Lightweight Transformer-Based Visual-Inertial Odometry for Robotics

Wenhui Wei, Yangfan Zhou, Yimin Hu, Zhi Li, Sen Wang, Xin Liu, Jiadong Li

Overview

Table of Contents

• Setup
• Data Preparation
• Download Pretrainined Models
• Evaluation
  • Pose Results and Visualization
  • Pose Evaluation
  • Depth Evaluation
  • Running Time Evaluation

Setup

• Create conda environment
• Install torch==1.12.1, torchvision==0.13.1, timm==0.4.12

Data Preparation

Please refer to Visual-Selective-VIO to prepare your data.

Download Pretrainined Models

Please download pretrained models and place them under pretrain_models directory.

Evaluation

Pose Results

python ./evaluation/eval_odom.py

Pose Evaluation

python ./evaluation/evaluate_pose_vo.py
Please modify '--data_path' in the options.py file to specify your dataset path. Additionally, update the pose embedding data type to float16 in PositionalEncodingFourier function within the depth encoder.py file.  In addtion, comment out the fully connected (FC) layer in the pose_encoder.py.

python ./evaluation/evaluate_pose_vio.py
Please modify '--data_path' in the options.py file to specify your dataset path. Additionally, update the pose embedding data type to float16 in PositionalEncodingFourier function within the depth encoder.py file.

Depth Evaluation

python ./evaluation/evaluate_depth.py
Please modify '--data_path' in the options.py file to specify your dataset path, and  comment out the IMU data reading process in the in the mono_dataset.py file.

Running Time Evaluation

python ./evaluation/evaluate_timing.py
Please modify '--data_path' in the options.py file to specify your dataset path. Additionally, update the pose embedding data type to float16 in PositionalEncodingFourier function within the depth encoder.py file.

Citation

The manuscript related to this work is currently under review.

Reference

[1] Lite-Mono, [2] ViTA