CAIRO Lab PhD candidate Lorin Achey is the lead author of a new paper accepted to the journal Autonomous Robots (AuRo):

Robust Robotic Exploration and Mapping Using Generative Occupancy Map Synthesis
by Lorin Achey, Alec Reed, Brendan Crowe, Bradley Hayes, and Christoffer Heckman.

This work presents a powerful new approach for autonomous robotic exploration that leverages generative diffusion models to enhance mapping in real time. Building on the SceneSense framework, the paper introduces a probabilistic map fusion technique that allows robots to synthesize plausible geometry in occluded or unseen regions of the environment—enabling faster, more reliable exploration and navigation.

Key contributions include:

• A real-time deployment of diffusion-based occupancy prediction on a Boston Dynamics Spot quadruped.
• A probabilistic map update rule that fuses predicted and observed occupancy, improving map fidelity and robustness.
• Demonstrated improvements of up to 75% in map realism and significant gains in exploration efficiency across challenging indoor environments.

The paper’s results (see Figures 5–7, pp. 10–11) show how SceneSense allows robots to “fill in” occluded geometry under and around themselves, enabling autonomous traversal through previously untraversable regions:contentReference[oaicite:1]{index=1}.

🔗 Project page: https://arpg.github.io/scenesense/

This work was supported by the National Science Foundation (Award #2339328) and continues the lab’s collaboration with the Autonomous Robotics & Perception Group (ARPG) at CU Boulder.