Legged manipulation platforms offer advantages over wheeled platforms, being able to move in difficult terrain (stairs, search and rescue scenario), adapt the base to ease the manipulation task and so on. In this project we want to build on the capabilities of our platform to efficiently detect a door handle, grasp it and open the door. Previous work has been done at our lab for detecting the door and identifying its parameters in addition to opening the door itself [1]. This project would consist of enhancing the existing elements, developing the remaining components, and finally integrating them into the pipeline.
References:
[1] Sleiman, Jean-Pierre et al. “A Unified MPC Framework for Dynamic Locomotion and Manipulation”, RAL 2021.
[2] He, Kaiming et al. “Mask R-CNN”
[2] https://github.com/facebookresearch/detectron2
Legged manipulation platforms offer advantages over wheeled platforms, being able to move in difficult terrain (stairs, search and rescue scenario), adapt the base to ease the manipulation task and so on. In this project we want to build on the capabilities of our platform to efficiently detect a door handle, grasp it and open the door. Previous work has been done at our lab for detecting the door and identifying its parameters in addition to opening the door itself [1]. This project would consist of enhancing the existing elements, developing the remaining components, and finally integrating them into the pipeline.
References:
[1] Sleiman, Jean-Pierre et al. “A Unified MPC Framework for Dynamic Locomotion and Manipulation”, RAL 2021.
- Literature review
- Training a neural network for door handle detection based on a Mask R-CNN [2]
- Developing a controller to accurately grasp the handle by integrating force-torque sensor data.
- Using the force-torque sensor to enhance the door-opening controller developed in [1].
- Literature review - Training a neural network for door handle detection based on a Mask R-CNN [2] - Developing a controller to accurately grasp the handle by integrating force-torque sensor data. - Using the force-torque sensor to enhance the door-opening controller developed in [1].
- Excellent C++ programming skills & some knowledge of ROS
- Good understanding of Robot Dynamics & Computer Vision
- Autonomy and dedication
- Excellent C++ programming skills & some knowledge of ROS - Good understanding of Robot Dynamics & Computer Vision - Autonomy and dedication