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/

Zendy Plus

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

Zendy Tools

Zendy Open

Choosing an appropriate representation of the environment for the underlyingdecision-making process of the reinforcement learning agent is not alwaysstraightforward. The state representation should be inclusive enough to allowthe agent to informatively decide on its actions and disentangled enough tosimplify policy training and the corresponding sim2real transfer. Given thisoutlook, this work examines the effect of various representations inincentivizing the agent to solve a specific robotic task: antipodal and planarobject grasping. A continuum of state representations is defined, starting fromhand-crafted numerical states to encoded image-based representations, withdecreasing levels of induced task-specific knowledge. The effects of eachrepresentation on the ability of the agent to solve the task in simulation andthe transferability of the learned policy to the real robot are examined andcompared against a model-based approach with complete system knowledge. Theresults show that reinforcement learning agents using numerical states canperform on par with non-learning baselines. Furthermore, we find that agentsusing image-based representations from pre-trained environment embeddingvectors perform better than end-to-end trained agents, and hypothesize thatseparation of representation learning from reinforcement learning can benefitsim2real transfer. Finally, we conclude that incentivizing the staterepresentation with task-specific knowledge facilitates faster convergence foragent training and increases success rates in sim2real robot control.

State Representations as Incentives for Reinforcement Learning Agents: A Sim2Real Analysis on Robotic Grasping