PiPER active scanning
Active vision for nature-surveying robots
I built an active-vision pipeline that deploys the arm to inspect targets obscured from a nature-surveying robot’s base camera.
View project repositoryView project posterWhat I built
What I built
I developed a ROS 2 pipeline for open-label target detection, mask tracking, depth projection, next-best-view planning and target-only RGB-D reconstruction. The intended surveying workflow starts with the base camera, deploys the arm for a clearer view, then repositions the base between scans if the target remains obscured. This repository provides the arm-level perception, planning and capture pipeline; the wider platform coordinates base repositioning.
Capabilities
- Open-label RGB-D perception with GroundingDINO and SAM2
- Occlusion-aware next-best-view planning with Tesseract and cuRobo
- Capture validation and target-only multi-view reconstruction












































