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

Robots often need to learn the human's reward function online, during thecurrent interaction. This real-time learning requires fast but approximatelearning rules: when the human's behavior is noisy or suboptimal, currentapproximations can result in unstable robot learning. Accordingly, in thispaper we seek to enhance the robustness and convergence properties of gradientdescent learning rules when inferring the human's reward parameters. We modelthe robot's learning algorithm as a dynamical system over the human preferenceparameters, where the human's true (but unknown) preferences are theequilibrium point. This enables us to perform Lyapunov stability analysis toderive the conditions under which the robot's learning dynamics converge. Ourproposed algorithm (StROL) uses these conditions to learn robust-by-designlearning rules: given the original learning dynamics, StROL outputs a modifiedlearning rule that now converges to the human's true parameters under a largerset of human inputs. In practice, these autonomously generated learning rulescan correctly infer what the human is trying to convey, even when the human isnoisy, biased, and suboptimal. Across simulations and a user study we find thatStROL results in a more accurate estimate and less regret than state-of-the-artapproaches for online reward learning. See videos and code here:https://github.com/VT-Collab/StROL_RAL

StROL: Stabilized and Robust Online Learning from Humans