The intelligent robotics comes in forms of mechanical design (i.e. embedded feedback mechanism into mechanical systems to reduce the complexity in control), interactive sensing, and artificial intelligence.
|Deep Learning Platform for Pick-and-Place|
The aim of this project is to explore autonomous and adaptive robotic pick-and-place through deep learning. Recent research by Google has demonstrated the possibility of training a robotic learning system with hand-eye-coordination during grasping task with significantly improved success rate. Yet to be validated, this approach could provide the flexibility on the extensive visual calibration and object recognition during usual robotic pick-and-place tasks. In human pick-and-place actions, there's another critical component that involves the dexterity of human hand and fingers, which enables the effective manipulation of objects. This is what we aim at exploring in this project, where a similar robotic setup is presented with a more advanced gripper with optional operation modes for adaptive grasping.
|as_DeepClaw: Grand Master of the Arcade Claw Machine Game|
as_DeepClaw is intended to be setup as a robotic grasping platform for deep learning based research and development. The robot is configured to perform similar tasks like the arcade game of claw crane. The goal of the game is for the player to control the crane in the horizontal x-y plane to determine an optimal coordinate to drop the claw in the vertical z axis. While descending, the claw will close while approaching the bottom of a transparent, enclosed cabinet filled with stuffed toys as rewards. If successful, one (or multiple if lucky) reward will be picked up by the closing claw, and deliver the reward to the lucky (or skilled) player by opening the claw above a drop hole. This is a very interesting game that’s been very popular in arcade game studios since the very beginning.
|Hyb-Ro Hand for Robotic Grasping|
To be continued...
|Multi-Axis Desktop CNC Machine|
Aimed at developing a multi-axis (4~6) CNC machines to the size factor of desktop usage, including the mechanical design, machine fabrication, and software integration, preferably with modularity introduced to the system design where certain axis groups could be reconfigured or replaced as a modular.
|Safety Issues in Human-Robot Interactions|
To be continued...
- (2017) Wan, F., and Song, C.*,
DeepClaw: a Robotic Learning Setup for Object Grasping using Arm-and-Gripper Robots,
Frontiers in Robotics and AI (Submitted).
- (2017) Wan, F. and Song, C.*,
Logical Learning Through a Hybrid Neural Network with Auxiliary Inputs,
arXiv: 1705.08200 [cs.AI].
- Wan, F., Jiang, W., Ong, B. and Song, C.*,
DeepClaw: A Universal Grasping Setup for Robotic Learning,
IEEE International Conference on Robotics and Automation (IEEE ICRA), Brisbane, QLD, Australia, May 21 – 25, 2018 (Submitted).
- Chen, Y., Le, S., Tan, Q., Lau, O., Wan, F. and Song, C.*,
A Lobster-inspired Robotic Glove for Hand Rehabilitation,
IEEE International Conference on Robotics and Automation (IEEE ICRA), Singapore, May 29 – June 03, 2017.