How BOTIX Improves Perception Robustness for Autonomous Navigation in Unpredictable Waters

Perception provides the essential input for decision-making and control in autonomous navigation, directly influencing whether a vessel can complete its voyage safely, cooperatively and efficiently.

Compared with road environments, however, waterways are inherently unpredictable. Complex and constantly changing conditions place significantly greater demands on perception-system robustness. Enabling autonomous vessels to see clearly, detect farther and perceive accurately in unstructured environments is therefore a major focus of BOTIX’s perception technology development.

Why Is Robust Perception So Difficult on the Water?

Waterborne and road environments differ fundamentally. Even as automotive perception technologies mature, they cannot simply be transferred to vessels.

Dynamic and unpredictable water conditions introduce more uncontrollable variables

Waves, currents and tides constantly alter the water surface. Vessel motions, including heave, pitch and roll, can also distort sensor data.

Unlike roads, waterways do not have fixed references such as lane markings. Navigable areas must instead be determined using factors such as water depth, restricted areas and navigation rules, making environmental modelling considerably more complex.

Water can also behave like a constantly changing mirror. Glare in bright sunlight and reflected lights at night may obscure critical targets and reduce visual perception performance.

Water amplifies the limitations of individual sensors and complicates sensor coordination

Each sensor type has its own strengths and limitations. In road vehicles, multiple sensors are commonly combined to provide complementary capabilities and establish a reliable perception network.

In marine environments, however, sensors face additional sources of interference. LiDAR may be affected by reflections and scattering from the water surface, resulting in noise and false detections. Cameras can be degraded by glare, refraction, surface ripples and low-light conditions. Millimeter-wave radar may also be affected by sea clutter and multipath reflections.

As a result, the expected complementary benefits of multisensor perception can be weakened by highly dynamic conditions and conflicting sensor information. This makes already demanding tasks such as data fusion, spatial-temporal alignment and environmental modelling even more complex.

Targets on the water exhibit highly variable motion patterns

Vessel motion is strongly coupled across six degrees of freedom, causing targets within sensor data to shake, scale or deform significantly.

Waves themselves are also dynamic sources of interference. They can make the apparent motion of a detected target differ from its actual movement, increasing the difficulty of target tracking and motion prediction.

BOTIX’s System-Level Approach: From Detection to Reliable Understanding

To address the uncertainty of waterborne environments, BOTIX has developed a multimodal sensor-fusion solution based on its experience in maritime autonomy.

The solution is designed to detect targets clearly, classify them accurately, maintain stable tracks and generate reliable motion predictions, supporting safe and dependable autonomous navigation.

Redundant sensor coverage provides a safety foundation

BOTIX’s perception solution incorporates multiple LiDAR units, cameras and millimeter-wave radars to create a multidimensional sensing architecture combining active detection with passive imaging.

Different sensor types respond differently to interference from waves, floating objects, reflections, spray and fog. This redundant architecture helps the system retain essential perception capabilities if an individual sensor fails or is temporarily degraded.

Noise-resistant algorithms reduce environmental interference

Point-cloud downsampling, motion compensation and haze and fog filtering help reduce interference caused by vessel motion and atmospheric conditions.

Image enhancement, reflection suppression and low-light optimization improve target visibility in strong sunlight, rain, fog and nighttime environments. Millimeter-wave radar provides additional resilience in adverse weather and can continue delivering target motion vectors in heavy rain and dense fog.

Through comprehensive data preprocessing, the system improves resistance to environmental interference and strengthens perception robustness across a broad range of operating and weather conditions.

Multimodal BEV fusion overcomes the limitations of individual sensors

BOTIX uses a multimodal BEV perception model to fuse heterogeneous data from visual cameras, LiDAR point clouds, thermal images and millimeter-wave radar point clouds at the feature level.

By integrating these inputs within a bird’s-eye-view representation, the system creates a unified understanding of the surrounding environment. This approach makes full use of each sensor’s strengths, maximizes redundancy and information complementarity, and establishes a more streamlined end-to-end perception pipeline within a common spatial framework.

The system also combines conventional detection algorithms with deep-learning models. This hybrid approach enables it to handle known sources of interference while adapting to unfamiliar operating conditions, balancing stability with generalization capability.

Perception is not an isolated function. It is a system-level engineering discipline encompassing hardware selection, software architecture, precise calibration, algorithm iteration, real-time optimization and redundant safety design.

BOTIX’s perception solution has been deeply integrated into the perception-decision-control loop of its autonomous navigation system, serving as the system’s central source of environmental understanding.

Whether on land or water, environmental perception still faces challenges that have yet to be fully resolved. There are no shortcuts in technology development. BOTIX will continue advancing coordinated innovation across its full technology stack, exploring new possibilities in marine perception and helping vessels navigate complex waters with greater safety, stability and confidence.