Research paper

OptLayer · Projecting robot actions into constraints

An optimization layer changes a neural policy’s proposed action to a nearby action satisfying chosen constraints. The abstract reports reaching tasks in simulation and on a real robot.

Reviewed 2026-11-05 · Source and bibliography checked

Bibliographic details and the editorial overview were checked against the original source. Access conditions and editions are identified with the source.

Original title: OptLayer - Practical Constrained Optimization for Deep Reinforcement Learning in the Real World
Publisher: IEEE
Authors: Tu-Hoa Pham, Giovanni De Magistris, Ryuki Tachibana
DOI: 10.1109/icra.2018.8460547
Conference date: 2018-05
Source language: en
Edition: 2018 IEEE International Conference on Robotics and Automation (ICRA)
Access: Publisher metadata and open author manuscript
Publication status: Published

Source overview

An optimization layer changes a neural policy’s proposed action to a nearby action satisfying chosen constraints. The abstract reports reaching tasks in simulation and on a real robot.

Use and review

Abstract and bibliographic review only. Guarantees depend on model, uncertainty and constraint assumptions; no ISO13849 safety function or complete installation was validated.

Sources & editions

OptLayer - Practical Constrained Optimization for Deep Reinforcement Learning in the Real World ↗

IEEE · 2018 IEEE International Conference on Robotics and Automation (ICRA)

Archivev2abstract/authors/history checked; IBMprimaryauthorpublicationpage separately identifies ICRA2018work, with repositorydate2018-09-10. Publisher-formatted manuscript was not compared.