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

This paper is an initial endeavor to bridge the gap between powerful DeepReinforcement Learning methodologies and the problem of exploration/coverage ofunknown terrains. Within this scope, MarsExplorer, an openai-gym compatibleenvironment tailored to exploration/coverage of unknown areas, is presented.MarsExplorer translates the original robotics problem into a ReinforcementLearning setup that various off-the-shelf algorithms can tackle. Any learnedpolicy can be straightforwardly applied to a robotic platform without anelaborate simulation model of the robot's dynamics to apply a differentlearning/adaptation phase. One of its core features is the controllablemulti-dimensional procedural generation of terrains, which is the key forproducing policies with strong generalization capabilities. Four differentstate-of-the-art RL algorithms (A3C, PPO, Rainbow, and SAC) are trained on theMarsExplorer environment, and a proper evaluation of their results compared tothe average human-level performance is reported. In the follow-up experimentalanalysis, the effect of the multi-dimensional difficulty setting on thelearning capabilities of the best-performing algorithm (PPO) is analyzed. Amilestone result is the generation of an exploration policy that follows theHilbert curve without providing this information to the environment orrewarding directly or indirectly Hilbert-curve-like trajectories. Theexperimental analysis is concluded by evaluating PPO learned policy algorithmside-by-side with frontier-based exploration strategies. A study on theperformance curves revealed that PPO-based policy was capable of performingadaptive-to-the-unknown-terrain sweeping without leaving expensive-to-revisitareas uncovered, underlying the capability of RL-based methodologies to tackleexploration tasks efficiently. The source code can be found at:https://github.com/dimikout3/MarsExplorer.

MarsExplorer: Exploration of Unknown Terrains via Deep Reinforcement Learning and Procedurally Generated Environments