Imagination-Driven Partially-Observable Task and Motion Planning
Published:
Mani-Imagine is a Task and Motion Planning Pipeline that aims to look for hidden targets in densely cluttered scenes. It is modeled as a Partially-Observable Markov Decision Process(POMDP) driven by the imagination capabilities of SOTA 3D Reconstruction Models and Vision-Language Models.
Features
- Partially-Observable Task and Motion Planning
- Can look for target objects in densely cluttered tabletop and shelf scenarios for now. More environments will be added later.
Technologies Used
- TAMP
- 3D Reconstruction Models
- Vision Language Models
NOTE: This project is under development and being finalized for upcoming conferences.