baxter-flowers / promplib

Python implementation of Probabilistic Motor Primitives including a ROS overlay to be used with JointTrajectory, JointState and RobotTrajectory, RobotState (Move It) messages
GNU Lesser General Public License v3.0
25 stars 11 forks source link

Issue with accounting for via points in TaskProMP #4

Open karthikm-0 opened 3 years ago

karthikm-0 commented 3 years ago

Hi,

Thank you for providing this very useful ProMP library! I had a question while implementing a TaskProMP. I am unable to have the provided demonstration take via points into account, and as a result, the generated trajectory is exactly the same as the demonstration. I was looking into the implementation details and noticed that the update step to account for each via point does not update the mean or covariance matrices - the addition term does not have an effect (lines 268-270 in promp.py), which explains why the generated trajectory is the same. I was wondering if you'd have any thoughts on how to fix this.

Thanks!

ymollard commented 3 years ago

You might meet incompatibilities coming from Python 3 or recent numpy/scipy versions. This library has not been maintained for 4 years and would honestly need a refresh for anyone willing to inspire from it.

Maybe test with some python/numpy/scipy versions that were alive in 2017 to observe if your issue still occurs. Comparing with conditioning equations from the original paper could also help. Also provide a test snippet.

karthikm-0 commented 3 years ago

Hi @ymollard thanks for bringing up the point about the Python/Numpy/Scipy versions. I've fixed the issue now using a consistent set of versions.