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

Simulation-to-real transfer is an important strategy for making reinforcementlearning practical with real robots. Successful sim-to-real transfer systemshave difficulty producing policies which generalize across tasks, despitetraining for thousands of hours equivalent real robot time. To address thisshortcoming, we present a novel approach to efficiently learning new roboticskills directly on a real robot, based on model-predictive control (MPC) and analgorithm for learning task representations. In short, we show how to reuse thesimulation from the pre-training step of sim-to-real methods as a tool forforesight, allowing the sim-to-real policy adapt to unseen tasks. Rather thanend-to-end learning policies for single tasks and attempting to transfer them,we first use simulation to simultaneously learn (1) a continuousparameterization (i.e. a skill embedding or latent) of task-appropriateprimitive skills, and (2) a single policy for these skills which is conditionedon this representation. We then directly transfer our multi-skill policy to areal robot, and actuate the robot by choosing sequences of skill latents whichactuate the policy, with each latent corresponding to a pre-learned primitiveskill controller. We complete unseen tasks by choosing new sequences of skilllatents to control the robot using MPC, where our MPC model is composed of thepre-trained skill policy executed in the simulation environment, run inparallel with the real robot. We discuss the background and principles of ourmethod, detail its practical implementation, and evaluate its performance byusing our method to train a real Sawyer Robot to achieve motion tasks such asdrawing and block pushing.

Simulator Predictive Control: Using Learned Task Representations and MPC for Zero-Shot Generalization and Sequencing