CS 5970-001: Main Class Project
This year's class project will be focused on the use of machine
learning methods to enable the robots to learn how to understand the
world around them and to act appropriately to solve tasks. In
particular, we are interested in how robots can use their experience
with the world to acquire sensing, planning and acting skill.
For our project, we will be using Karpal, a Baxter-type robot from
Rethink Robotics. We also have access to Yatima, a mobile
platform that is built on top of a Segway.
Possible project foci include:
- Learning how best to grasp an object, based on a variety of
properties, including: location, size, shape, color (supervised
- Learning general grasping skills, based on details of shape as
captured in a pointcloud (supervised learning)
- Putting sequences of actions together to accomplish a larger
task (reinforcement learning). Interesting tasks are those
that require different sequences of actions, depending on the
configuration of objects (including regrasping or repositioning
- Learning how to do a cooperative task with a human
- Learning how to do a cooperative task with another robot
The grading details are available in the syllabus.
Possible milestones include:
- Using visual input, reach to grasp a simple object in an uncluttered workspace.
- Reach to grasp a complex object in an uncluttered workspace.
- Use vision to select which object to grasp.
- Push an object from one location to another.
- Place a grasped object inside of a container or on a surface.
- Open a drawer or a door.
- Pour items from one container to another.
- Transfer an object from one hand to the other.
- Drive so that a carried object is graspable.
- Fully integrate SMRF trees and other learning algorithms into
our environment (including translation of sensory data into
appropriate input form and translation of outputs to robot actions).
- In large part, we will be implementing using a mixture of
python and C++
- Documentation will be done using doxygen
- See the syllabus for
details on expectations for proper academic conduct (including
use of the code of others).
- Grading of a component will be triggered by a
presentation/demonstration of the component in class. Only
code submitted to the subversion tree will
andrewhfagg [[at]] gmail.com
Last modified: Mon Jan 18 15:00:13 2016