Lucas Manuelli

Spartan Feature Requests

1

General issue for tracking feature requests. Break out into more specific issues as needed. - [ ] Publish status message from controller.For example `kuka_plan_runner` could publish the time to go...

Standardize messages for kinematic trajectory planning.

1

Following up on the conversation with @gizatt @peteflorence @avalenzu @hongkai-dai at TRI earlier today. I have created a new [repo](https://github.com/manuelli/drake_kinematic_trajectory_planning_msgs) to standardize the messages for interacting with drake's kinematic trajectory...

I believe we have more conservative values in kuka_driver.cc to avoid triggering estops on the robot. We should just apply these in the urdf as well.

Otherwise the driver is constantly being toggled on/off and produces corrupted images

director freezes when moving pointcloud

4

The symptom is that if you have a pointcloud visualized, add a child frame to it with `vis.addChildFrame()` and then try to move that child frame, director freezes, in fact...

- [x] Implement abstract Q-Learning Class - [x] SARSA, Q-Learning, should all be subclasses Methods to implement - [x] SARSA, see #32 - [ ] policy search / policy gradient,...

SARSA

1

- [x] Implement SARSA update - [x] Test it for convergence with a known policy, i.e. our simple controller - [x] Implement full SARSA policy updates

Start thinking about policy gradient methods. What would we need to implement them? What should be the parametric form of the policy?