Commercial grade, faster processing, lower memory, higher accuracy, and greater robustness

• Kudan’s proprietary visual SLAM software has been extensively developed and tested for use in commercial settings.
• Open source and other commercial algorithms struggle in many common use cases and scenarios.
• KudanSLAM achieves much faster processing time, higher accuracy and more robustness in dynamic situations.

Reference datasets are available here (Dataset 1 – a visual sequence targeting indoor robots, Dataset 2 – a visual sequence targeting AR/VR)

Overcome typical Lidar localization and mapping problems with higher accuracy, reduced map size, and lower latency

• Leverages our deep expertise in visual SLAM to reach groundbreaking performance, including 1cm-level accuracy, and greater than 300x lighter map size over raw Lidar data.
• Almost zero pose estimation latency when combining with an IMU.
• Compatible with many different types of Lidar technology such as, spinning, solid state, and prizm based.
• Ability to correct motion blur in the Lidar data stream while mapping resulting in sharper point clouds and improved post AI process such as object recognition
• Reference dataset is available here
  (Dataset – a dataset targeting autonomous driving, outdoor robotics: coming soon)

Artisense’s GN-net relocalization: AI-enabled (re)localization that is incredibly robust to scenery changes

• Overcomes the challenge of localization, and mapping a vehicle or machine when the scenery or conditions change over time.
• Uses “deep features” identified by AI for matching correct locations even when their appearance is different.
• SLAM mapping can update automatically to include any changes that are detected.

Integrate all the sensor information into one system rather than just “filtering” them with accurate time synchronization

• Common sensor fusion approach is a “loose-coupling” of sensors, while Kudan and Artisense use “tight-coupling”.
• The Kudan group has developed the only system that uses tight-coupling of all sensors including camera, Lidar, GNSS and IMU.
• Problematic issues of proper time synchronization are solved with micro-second level synchronization between all sensors.
• Unparalleled robustness and accuracy for a wide range of use cases.

Enabling LiDAR OEMs and users to achieve better extrinsic calibrations and correct motion blur using our extensive computer vision capability leading to more accurate maps and improved recognition performance

• Lack of accurate extrinsic calibrations and motion blur cause highly visible degradation to LiDAR point clouds in all applicable use cases.
• Accurate localization and understanding of vehicle / machine ego-motion can provide better extrinsic calibrations between sensors and sharper point cloud maps even with high-speed movement or fast rotation.
• Applicable to more than just SLAM software, the improved calibration and point clouds are also applicable to image recognition, resulting in increased safety and system reliability.

Practical functionality for large-scale map management

• Simple to use map handling such as loading, merging, splitting, saving, and unloading.
• Multiple maps can be loaded to cover the same area, allowing unmatched performance against changing scenery.
• Maps can be loaded and unloaded into Kudan software while running allowing for unlimited scale maps.

Unprecedented speed and other performance gains at low power through processor and sensor optimization

• Kudan has been proactively working with sensor and processor players to optimize KudanSLAM algorithms.
• Provides unparalleled speed, robustness, accuracy and power consumption when combined with specific processors / sensors.