1st Project
Development of an algorithm for creating sphere representation of complex objects which allows faster collision checking.
Purpose
Motion planning involves a lot of collision checking.
Reduction of the collision checking calculation is important for the real-time perception-based motion planning.
Therefore we would like to simplify meshes of robots into spheres.
Current Method
Current method takes only box shapes as inputs. It cannot deal with the objects of general shapes.
Output of the current method is just a series of spheres (array) which are not efficient for the collision checking calculation.
Approach
New method takes general shapes of objects as inputs.
Output of the new method is a hierarchical tree structure of spheres (sphere-tree) which is efficient for the collision checking calculation.
Hierarchical tree structure of spheres (sphere-tree)
Number of Calculations for Collision Checking (Ex.)
Result
Bunny
Elbow Flex Link
2ND PROJECT
Joystick teleoperation of the end-effector of Fetch
PURPOSE
There are three kinds of control for manipulation: Pose(position), twist(velocity), wrench(torque).
It is necessary to enable a joystick to control the end-effector to check if the controller is working fine.
APPROACH
Reference frame (body: blue, End-effector: red)
Mapping linear and angular velocity to key #1, #2, and tilting motions.
Converting reference frame using key #11 and #9.
Using KDL library to solve 7 DOF arm kinematics.
RESULT
Code can be found at Fetch Robotics Github as an open source. Fetch users are now able to control the end-effector using the joystick.