Any update about the drake panda driver? Looking forward to it : )

Should $a$ here also be a vector? Since for different state $s_i$ the optimal action can be different.

Hi, I have a question about the pure torque control mode of real iiwa robot. If the desired position is setted as the current position of robot, then it seems that iiwa can be used in pure torque control mode through \tau_{ff} (Maybe with compensation of gravity by the robot). However, in the problem 7.3, there is a comment by terry-suh saying that for real iiwa robot, this cannot be done. I am a bit curious about why the pure torque control cannot be done in real iiwa robot, given that Franka Panda provides pure torque control API as its documentation says so.

I think that usually the state \x will have a nonlinear relationship with \alpha, the policy parameter. Therefore, write \x = f(\alpha) + \beta maybe more appropriate. This will add a jacobian matrix ∂f/∂\alpha in the result.

Hi, I have a question about which rendering engine is used for this simulaiton video. Is blender being used? Same question goes for the video in the begining of this post. https://medium.com/toyotaresearch/drake-model-based-design-in-the-age-of-robotics-and-machine-learning-59938c985515. It looks like that the video in the beginning of the above post is rendered with blender.