OMRON SINIC X to Present Robot Learning Research at CoRL 2026
Two research papers from OMRON SINIC X have been accepted for presentation at the CoRL 2026 conference. The findings focus on adapting language models for robot control and real-time failure detection.
OMRON Corporation subsidiary OMRON SINIC X will present two research papers at the Conference on Robot Learning (CoRL) 2026, scheduled for November in Austin, Texas. The conference is a significant event in the robotics field, with an acceptance rate of approximately 33 percent for this year's cycle.
The first paper, titled "CLAP: Direct VLM-to-VLA Adaptation via Language-Action Grounding," introduces a method to adapt pre-trained vision-language models for robot control without requiring architectural modifications. The system achieved a 90.8% success rate on the LIBERO manipulation benchmark, demonstrating that integrating natural language action descriptions can optimize robot performance.
The second study, "Online Robot Manipulation Error Detection and Description," presents the STREAM-FAIL framework. This system enables robots to detect and describe manipulation failures in real time using a memory-bank approach. The team developed an automated pipeline to convert robot trajectory data into temporal grounding annotations, which reportedly improved failure localization by 73% relative to existing methods.
By open-sourcing the research methodologies and datasets, OMRON SINIC X aims to support further development in the field of autonomous robot reliability and interaction.