🤖 RoboSense Track 3:
Sensor Placement

Official Baseline Implementation for Track 3

Based on Place3D -- "Is Your LiDAR Placement Optimized for 3D Scene Understanding?"
(https://github.com/ywyeli/Place3D)

🏆 Prize Pool: $2,000 USD for Track 3 Winners

Challenge Overview

Track 3: Sensor Placement challenges participants to design LiDAR-based 3D perception models, including those for 3D object detection, that can adapt to diverse sensor placements in autonomous systems.

Participants will be tasked with developing novel algorithms that can adapt to and optimize LiDAR sensor placements, ensuring high-quality 3D scene understanding across a wide range of environmental conditions, such as weather variances, motion disturbances, and sensor failures.

🎯 Objectives

To be added.

Competition Details

• Venue: IROS 2025, Hangzhou (Oct 19-25, 2025)
• Registration: Google Form (Open until Aug 15)
• Contact: robosense2025@gmail.com

🏆 Awards

Prize	Award
🥇 1st Place	$1000 + Certificate
🥈 2nd Place	$600 + Certificate
🥉 3rd Place	$400 + Certificate
🌟 Innovation Award	Cash Award + Certificate
Participation	Certificate

📊 Official Dataset

The track 3 dataset is prepared using data and settings from the Place3D framework.

Dataset Statistics (Phase 1)

In Phase 1, we provide a trainval dataset consisting of camera images from six views and LiDAR data from four different placements. You may choose to train your model on one placement and validate it on others, or design methods that leverage all placements to achieve better generalization.

The dataset contains 200 scenes (8,000 frames) in total, split into 125 scenes (5,000 frames) for training and 75 scenes (3,000 frames) for validation.

All LiDAR data from different placements are collected simultaneously, and they share the same metadata and annotation files. You can create symbolic links pointing to the desired data path, in order to conveniently switch between LiDAR placements.

File name	Description	Split
track3_phase1_lidar_blobs_only_placement_0	LiDAR data, placement Center	train, val
track3_phase1_lidar_blobs_only_placement_1	LiDAR data, placement Line	train, val
track3_phase1_lidar_blobs_only_placement_2	LiDAR data, placement Trapezoid	train, val
track3_phase1_lidar_blobs_only_placement_3	LiDAR data, placement Pyramid	train, val
track3_phase1_camera_blobs_only	Camera images of 6 views	train, val
track3_phase1_sensor_file_blobs	LiDAR (Center) + Camera data	train, val
track3_phase1_metadata	metadata including annotations (gt)	train, val
track3_phase1and2_map	maps	-

Dataset Statistics (Phase 2)

In Phase 2, we provide a test dataset that includes camera images from six views and LiDAR data from six different placements. All LiDAR data are organized in a single folder, so you do not need to switch paths when running tests.

Please use the Phase 1 trainval dataset to train and validate your model, and then evaluate its performance on the Phase 2 test set. See the following sections for detailed instructions on how to submit your results.

File name	Description	Split
track3_phase2_lidar_blobs_only_test_part	LiDAR data	test
track3_phase2_camera_blobs_only_test_part	Camera images of 6 views	test
track3_phase2_sensor_file_blobs_test_part	LiDAR + Camera data	test
track3_phase2_metadata	metadata without annotations (gt)	test
track3_phase1and2_map	maps	-

Baseline Performance (Phase 1)

To be added.

🚀 Quick Start

We provide a simple demo to run the baseline model.

1. Preparing conda env

The code is built with following libraries:

• Python >= 3.8, <3.9
• OpenMPI = 4.0.4 and mpi4py = 3.0.3 (Needed for torchpack)
• Pillow = 8.4.0 (see here)
• PyTorch >= 1.9, <= 1.10.2
• tqdm
• torchpack
• mmcv = 1.4.0
• mmdetection = 2.20.0
• numpy = 1.23

After installing these dependencies, please run this command to install the codebase:

cd projects/bevfusion
python setup.py develop

We also provide a Dockerfile to ease environment setup. To get started with docker, please make sure that docker is installed on your machine. After that, please execute the following command to build the docker image:

cd projects/bevfusion/docker && docker build . -t bevfusion

We can then run the docker with the following command:

docker run --gpus all -it -v `pwd`/../data:/dataset --shm-size 16g bevfusion /bin/bash

We recommend the users to run data preparation (instructions are available in the next section) outside the docker if possible. Note that the dataset directory should be an absolute path. Within the docker, please run the following command to clone our repo and install custom CUDA extensions:

git clone https://github.com/robosense2025/track3 && cd projects/bevfusion
python setup.py develop

You can then create a symbolic link data to the /dataset directory in the docker.

2. Prepare the dataset

Add dataset_utils path to your PYTHONPATH. Edit ~/.bashrc or ~/.zshrc:

export PYTHONPATH="$PYTHONPATH:/[YOUR_PARENT_FOLDER]/track3/projects/dataset_utils"

Apply the change:

source ~/.bashrc  # or ~/.zshrc

Please download the dataset from here. Our dataset is in nuScenes format, but requires our customized tools.

We typically need to organize the useful data information with a .pkl or .json file in a specific style for organizing annotations. To prepare these files, run the following command:

python tools/create_data.py nuscenes --root-path ./data/nuscenes --out-dir ./data/nuscenes --extra-tag nuscenes

After data preparation, you will be able to see the following directory structure (as is indicated in mmdetection3d):

mmdetection3d
├── mmdet3d
├── tools
├── configs
├── data
│   ├── nuscenes
│   │   ├── maps
│   │   ├── samples
│   │   ├── v1.0-test
|   |   ├── v1.0-trainval
│   │   ├── nuscenes_database
│   │   ├── nuscenes_infos_train.pkl
│   │   ├── nuscenes_infos_val.pkl
│   │   ├── nuscenes_infos_test.pkl
│   │   ├── nuscenes_dbinfos_train.pkl

3. Deploy

To be added.

Training

For LiDAR detector, please run:

torchpack dist-run -np 8 python tools/train.py configs/nuscenes/det/transfusion/secfpn/lidar/voxelnet_0p075.yaml

4. Evaluate the baseline

You will be able to run:

torchpack dist-run -np 8 python tools/test.py [config file path] pretrained/[checkpoint name].pth --eval [evaluation type]

For example:

torchpack dist-run -np 8 python tools/test.py configs/nuscenes/det/transfusion/secfpn/lidar/voxelnet_0p075.yaml pretrained/track3-baseline.pth --eval bbox

🎖️ Challenge Participation

Submission Requirements

For Phase 1 submission, run the following command:

torchpack dist-run -np 8 python tools/test.py [config file path] pretrained/[checkpoint name].pth --eval bbox --eval-options 'jsonfile_prefix=./results'

For example:

torchpack dist-run -np 8 python tools/test.py configs/nuscenes/det/transfusion/secfpn/lidar/voxelnet_0p075.yaml pretrained/track3-baseline.pth --eval bbox --eval-options 'jsonfile_prefix=./results'

After running, you will find a file named results_nusc.json in ./results. Zip this file and submit results_nusc.zip to Codabench.

For Phase 2 submission, in /track3/projects/bevfusionconfigs/nuscenes/default.yaml,

1. update line 289 to ann_file: ${dataset_root + "nuscenes_infos_test.pkl"}.

2. Comment out lines 207-211 that loads annotations (LoadAnnotations3D) in the test pipeline.

3. Comment out lines 247-249 that includes the keys of gt data (Collect3D).

Then, run the same command from Phase 1 after making the above changes.

📏 Evaluation Metrics

The evaluation metrics are same as nuScenes. (Do not install the official nuscenes-devkit, as it may cause compatibility issues.)

Timeline

• Registration: Google Form
• Phase 1 Deadline: August 15th
• Phase 2 Deadline: September 15th
• Awards Announcement: IROS 2025

🔗 Resources

• Challenge Website: robosense2025.github.io
• Track Details: Track 3 Page
• Track Dataset: HuggingFace Page
• Baseline Model: HuggingFace Page
• Related Paper: arXiv:2403.17009

📧 Contact & Support

• Email: robosense2025@gmail.com
• Official Website: https://robosense2025.github.io
• Issues: Please use GitHub Issues for technical questions

📄 Citation

If you use the code and dataset in your research, please cite:

@inproceedings{li2024place3d,
  title = {Is Your LiDAR Placement Optimized for 3D Scene Understanding?},
  author = {Ye Li and Lingdong Kong and Hanjiang Hu and Xiaohao Xu and Xiaonan Huang},
  booktitle = {Advances in Neural Information Processing Systems},
  year = {2024}
}

Acknowledgements

RoboSense 2025 Challenge Organizers

RoboSense 2025 Program Committee

---

🤖 Ready to sense the world robustly? Register now and compete for $2,000!

📝 Register Here | 🌐 Challenge Website | 📧 Contact Us

Made with ❤️ by the RoboSense 2025 Team