The cognition engine for embodied intelligence.
CerebrumFlow is a research group studying how machines perceive, reason, and act in the physical world. We connect perception, planning, and control into a single loop — and ask how the result can move from simulation into real robots, reliably and repeatably.
Our work sits at the intersection of robotic perception, motion planning, and learned control. We publish in the robotics and learning venues, release the datasets and benchmarks that make our results reproducible, and keep a long view toward machines that reason about their own actions.
Four threads, one loop.
Each area is its own line of inquiry, but none of them works alone — the interesting questions sit at the boundaries between them.
Perception
Scene understanding and predictive uncertainty — building representations a robot can reason over, not just label.
Learn more →Planning
Search and reasoning over uncertain representations, from long-horizon task plans to real-time motion.
Learn more →Control
Whole-body and reactive control from learned policies, with the stability guarantees real hardware demands.
Learn more →Sim-to-real
Transferring learned behaviour from simulation to hardware without the brittleness that usually follows.
Learn more →Selected publications
FlowBench: A Benchmark for Embodied Perceive–Plan–Act Systems
IEEE International Conference on Robotics and AutomationA Unified Token Interface for Perception, Planning, and Control
IEEE/RSJ International Conference on Intelligent Robots and SystemsLatent World Models for Contact-Rich Manipulation
Conference on Robot LearningPerceive–Plan–Act with Predictive Uncertainty in Cluttered Scenes
IEEE International Conference on Robotics and AutomationA small, senior team.
Six researchers with backgrounds across planning, control, and learning — plus the engineering needed to turn results into running systems.
Meet the team →Reproducible by default.
We release the benchmarks, datasets, and software behind our papers, so others can build on the same ground truth we do.
Browse the releases →