Embodied Neural Image Signal Processor For Robotic Manipulation

Vision-language-action models typically operate on RGB images produced by a fixed image signal processor, leaving the camera pipeline outside the learning and evaluation loop. We systematically examine the consequences of this overlooked design choice across five fundamental ISP dimensions: gain, sensor noise, color rendering, tonal mapping, and bit depth. Our analysis reveals that RAW-to-RGB processing materially shapes both action prediction and manipulation success, with different ISP dimensions exerting substantially different effects. Guided by these findings, we introduce RawVLA, a neural ISP that adaptively renders RAW observations for frozen VLA policies while concentrating its capacity on the imaging factors most relevant to embodied behavior. We further present RawVLA-Bench, a RAW-domain manipulation benchmark built upon LIBERO and RoboTwin2 to expose image processing as an explicit evaluation variable across clean and challenging light conditions. Experiments on RawVLA-Bench show that our method preserves performance under standard conditions while substantially improving robustness under degraded imaging, establishing adaptive RAW processing as an effective interface between physical cameras and embodied policies. Source code and data will be publicly available.