diff --git a/docs/index.html b/docs/index.html index 7ec0599..7851a1c 100644 --- a/docs/index.html +++ b/docs/index.html @@ -15,8 +15,7 @@ - + @@ -134,8 +133,7 @@
In order to showcase that ANDPs can work with arbitrary state spaces
+ 1, we learn the spiral motion but this time in joint space (7 degrees of freedom), this means that x ≡ xc =
+
+
+
+
+ and the output is the desired velocity profile ẋc that the joints should follow.
+ The data collection process of this experiment is similar to the one described above, with the only difference that this time on every timestep we collect the angle of every joint of the robot, as well as their corresponding angular velocities. The results showcase that we are able to learn the task even in the higher dimensional space of the joints.
+
+ θ0
+ θ1
+ ⋯
+ θ6
+
1 Formally, we should require that the controllable part of the state forms a Euclidean space, which holds for the reduced coordinates system (joint space). The derivations and discussion on this topic are out of the scope of this paper, and will be discussed in future work.
+ +- + | -
- + |