PR2 Hyperparms Question

[Olalekan Ogunmolu]

Hi there,

In the PR2 hyperparameters file, I noticed the initial state was initialized by a numpy array of 1X32 zeros.

I'm talking about this line of code here: https://github.com/cbfinn/gps/blob/master/experiments/pr2_example/hyperparams.py#L72-#L76

The robot has 7 arms and the conditions state in the config.py file is set to 1.

Whence comest 7 X 4 = 32? Or in contemporary lingo, why is this so? :)

Also, we initialize the first 7 elements  from the initial state of the joint angles of the robot and the 14th:23rd  elements of the state are populated from the flattened end_effector points but why do we leave the rest of the states at zero?

I'm not sure this is mentioned in the paper.

Lekan

[Chelsea Finn]

The state includes the joint angles and velocities (7x2=14) and the pose&velocity of the end effector, represented as 3 points in 3D (3x3x2=18).

The rest of the state is left to zero because we assume that the initial velocity of the arm is 0.

Chelsea

--
You received this message because you are subscribed to the Google Groups "gps-help" group.
To unsubscribe from this group and stop receiving emails from it, send an email to gps-help+unsubscribe@googlegroups.com.
To post to this group, send email to gps-...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/gps-help/0998931b-9e76-4f22-a0f1-4a3f6cbefe46%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Chelsea Finn]

​This is just for the controller initialization, not for trajectory optimization.​

Chelsea

On Mon, Dec 5, 2016 at 10:16 AM, Lekan <patl...@gmail.com> wrote:

Thank you! That makes sense.

I have one more question, please. In initializing the linear Gaussian controllers, you defined Ltt, the Hessian of loss with restect to the trajectory at  a single time step as

Ltt = np.diag(np.hstack([
	config['stiffness'] * np.ones(dU),
	config['stiffness'] * config['stiffness_vel'] * np.ones(dU),
	np.zeros(dX - dU*2), np.ones(dU)
	]))

Here: https://github.com/lakehanne/gps/blob/master/python/gps/algorithm/policy/lin_gauss_init.py#L51-#L57

Can please point me to the paper where the cost function equations and trajectory optimization equations are spelled out? I am not too sure why you have this term in the Hessian: np.zeros(dX - dU*2).

I am implementing the algorithm on a custom robot.

--

Lekan