Journal
Research, talks and company news.
Bpifrance supports Reflexion’s work on robotic execution
Bpifrance supports our work on the safety runtime that governs the execution of robotic policies inside Reflexion OS.
Read the announcement →
Reflexion Robotics is a member of NVIDIA Inception.
Research & journal
Our architecture and the convergence of agentic robotics
Reflexion has been working since 2025 on a system-level approach to embodied AI built around memory, hierarchical reasoning, callable capabilities, execution feedback and independent physical control.
Recent work across the robotics field increasingly exposes many of these same architectural primitives.
We look at that convergence, the public work that preceded it and what it means for Reflexion OS.
Read the article →A hierarchical architecture for embodied AI
Published in November 2025, our Founder & CTO’s preprint describes the separation between high-level planning, callable robot skills and independent real-time physical enforcement. The work forms part of the architectural foundation behind Reflexion OS.
Read the paper →From context engineering to robot memory
At GOSIM Hangzhou in September 2025, our Founder & CTO presented work connecting persistent context, memory and robot action. That work became one of the foundations of our approach to long-running physical and digital agents.
Read more →Skill Foundry at GOSIM Paris
In May 2026, our Founder & CTO presented Skill Foundry: an embodied agent that watches its own execution and develops new robot skills through an iterative learning process.
View the session →Reflexion at GOSIM Shenzhen
Our Founder & CTO will present “Train It or Code It? Let the Robot Decide” at GOSIM Shenzhen, continuing our work on agents that can choose between existing skills, code and learned behavior depending on the task.
Read more →
All articles
- Research
Our architecture and the convergence of agentic robotics
A system built around the task: memory, reusable capabilities and observed execution.
- Research
A hierarchical architecture for embodied AI
Our Founder & CTO’s paper on task planning, callable skills and independent physical enforcement.
- Research
From context engineering to robot memory
Connecting persistent context to robot action at GOSIM Hangzhou.
- Research
Skill Foundry at GOSIM Paris
An agent that observes its own behavior and develops new robot skills.
- Company
Reflexion at GOSIM Shenzhen
Join us to discuss how robots choose capabilities and how to evaluate their consequences.
- Company
Bpifrance supports Reflexion’s work on robotic execution
Bpifrance supports our work on the safety runtime for robotic policy execution.
- Company
Reflexion Robotics is a member of NVIDIA Inception
Reflexion Robotics is a member of NVIDIA Inception.
- Engineering
Adding modern AI to existing robots
Start with a task and the equipment you have.