The RoboSense Challenge 2025

👋 Welcome to Track #5: Cross-Platform 3D Object Detection of the 2025 RoboSense Challenge!

As robotics continues to advance, LiDAR-based 3D object detection has become a focal point in both academia and industry. However, most existing datasets and methods target vehicle platforms, overlooking quadrupeds and drones. This challenge, built on our benchmark, aims to:

1. Build on three platforms—vehicles, drones, and quadruped robots—to foster innovations in a unified perception framework;
2. Bridge geometric and data distribution disparities to achieve rapid model transfer and adaptation across platforms;
3. Lower annotation and deployment overhead, supporting collaborative sensing for heterogeneous robot teams in urban, disaster, and indoor scenarios.

Phase 1: Vehicle → Drone Adaptation

Duration: 15 June 2025 – 15 August 2025

Setup:

• Source platform: Vehicle LiDAR scans with 3D bounding-box annotations
• Target platform: Unlabeled Drone LiDAR scans

Ranking Metric: AP@0.50 (R40) for the Car class evaluated on Drone data

Phase 2: Vehicle → Drone & Quadruped Adaptation

Duration: 15 August 2025 – 15 September 2025

Setup:

• Source platform: Vehicle LiDAR scans with annotations
• Target platforms: Unlabeled Drone and Quadruped LiDAR scans

Ranking Metric: Weighted score combining:

• AP@0.50 (R40) for the Car class
• AP@0.25 (R40) for the Pedestrian class

(Scores computed across both Drone and Quadruped platforms.)

In this track, we adopt PV-RCNN as the base 3D detector, and leverage ST3D/++ as our baseline adaptation framework. Detailed environment setup and experimental protocols can be found in the Track5 GitHub repository .

Beyond the provided baseline, participants are encouraged to explore alternative strategies to further boost cross-platform performance:

• Treat the cross-platform challenge as a domain adaptation problem by improving pseudo-label quality and fine-tuning on target-platform data.
• Design novel data augmentation techniques to bridge geometric and feature discrepancies across platforms.
• Adopt geometry-agnostic 3D detectors, such as point-based architectures, that are less sensitive to platform-specific point-cloud characteristics.

Phase 1 Results

We provide the following resources to support the development of models in this track: