Published January 8, 2020 | Version v1
Poster Open

Locally Weighted CNMPs for Generating Flexible Action Sequences

Authors/Creators

  • 1. Bogazici University
  • 2. Ozyegin University
  • 3. Osaka University

Description

This is a sequel work which introduces locality prior to CNMPs to increase its representational power. Instead of using the average operator for the aggregation, we use weights that are generated by a Gaussian that is defined by the query point. This way, we can generate trajectories by combining local primitives. We call this model locally weighted CNMPs.

Files

Ahmetoglu-2020-BrainAndMind.pdf

Files (10.4 MB)

Name Size Download all
md5:55844018d94f9f9e6845de9e9fccc0e9
10.4 MB Preview Download

Additional details

Funding

European Commission
IMAGINE - Robots Understanding Their Actions by Imagining Their Effects 731761