Xiang Wang∗ and Peter Xiaoping Liu,∗,∗∗
NeRF, low-light environment, robotics mapping, multi-resolution hash encoding
Low-light environments pose a significant challenge for robots during task execution, severely limiting their perception capabilities, particularly in mapping. Under insufficient lighting conditions, robots often struggle to perform effectively, with the generated images typically appearing overly dim, exposing notable weaknesses and limitations in such scenarios. To address this issue, we propose an NeRF-based real-time mapping method for robots operating in low-light environments. By introducing a low-light field, the model effectively learns geometric and texture features under low-light conditions, enabling high-quality scene reconstruction comparable to that achieved under normal lighting conditions. Additionally, we leverage hierarchical hybrid representation to further accelerate the mapping process. Extensive experiments on multiple datasets and real-world robotic scenarios demonstrate that the proposed method significantly enhances mapping performance in low-light environments, showcasing outstanding effectiveness and practical applicability.
Important Links:
Go Back
