English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Model predictive control (MPC) is a popular approach for trajectoryoptimization in practical robotics applications. MPC policies can optimizetrajectory parameters under kinodynamic and safety constraints and provideguarantees on safety, optimality, generalizability, interpretability, andexplainability. However, some behaviors are complex and it is difficult tohand-craft an MPC objective function. A special class of MPC policies calledLearnable-MPC addresses this difficulty using imitation learning from expertdemonstrations. However, they require the demonstrator and the imitator agentsto be identical which is hard to satisfy in many real world applications ofrobotics. In this paper, we address the practical problem of trainingLearnable-MPC policies when the demonstrator and the imitator do not share thesame dynamics and their state spaces may have a partial overlap. We propose anovel approach that uses a generative adversarial network (GAN) to minimize theJensen-Shannon divergence between the state-trajectory distributions of thedemonstrator and the imitator. We evaluate our approach on a variety ofsimulated robotics tasks of DeepMind Control suite and demonstrate the efficacyof our approach at learning the demonstrator's behavior without having to copytheir actions.

GAN-MPC: Training Model Predictive Controllers with Parameterized Cost Functions using Demonstrations from Non-identical Experts