La IA física recopila datos del mundo real utilizando sensores, como cámaras, detección y medición de luz (LiDAR) y radar, para permitir que los modelos de IA implementados en sistemas autónomos y robots comprendan el espacio físico y así puedan desempeñarse en condiciones reales.

Sensor fusion is the process of combining data from multiple sensors to improve the accuracy and reliability of the information. By combining information from different types of sensors (such as LiDAR, radar, and cameras), a system can create a more complete picture of the environment. Sensor fusion can also reduce the impact of errors or failures in any individual sensor. This technology is used in various applications, including autonomous vehicles (AV) and robotics.

One of the most common applications of sensor fusion is the combination of various sensors mounted on a robotaxi including 3D point clouds from LiDARs and 2D images from front, side and rear cameras.

Los sensores son dispositivos que detectan y responden a estímulos físicos, como la luz, el calor y el sonido. Diferentes sensores utilizan distintas tecnologías para realizar sus funciones. Por ejemplo, los sensores LiDAR y de radar emplean láseres y ondas de radio para percibir su entorno, mientras que los sensores ultrasónicos utilizan ondas sonoras.

LiDAR (Light Detection and Ranging)

High accuracy, long-range, and fast data acquisition. LiDAR is ideal for mapping and obstacle avoidance in AV and robots.

RADAR (Radio Detection and Ranging)

Radar sensors can detect objects through various weather conditions and at long distances, making them ideal for applications such as collision avoidance and autonomous driving.

SONAR (Sound Navigation and Ranging)

Sonar sensors emit sound waves and measure the time it takes for the waves to bounce back after hitting an object. In robotics and automotive they can be used to determine the distance to the object.

Structured light

Structured light sensors use a 3D scanner to measure the 3D dimensions of an object. High resolution and accuracy make them suitable for 3D scanning and mapping applications.

Time-of-Flight (ToF) sensor

ToF sensors are used for measuring distance with depth sensing technology. They are fast and reliable, so they are a good choice for gesture recognition, object tracking, and robot navigation applications.

Stereo vision sensor

Stereo vision sensors use two (or more) sensors to simulate human binocular vision. High precision depth sensing, good spatial resolution, and low cost make these suitable for obstacle avoidance, 3D mapping, and robot navigation applications.

Ultrasonic sensor

Ultrasonic sensors are low-cost, easy to use, and can detect a wide range of materials, making them suitable for applications such as parking assistance and object detection.

Step 1: Overlay the data

• The first step is to calibrate and align all the sensor views, then overlay all of your data into a single view. This fused view gives your labeler more context, allowing it to tell what groups of point clouds represent.

Step 2: Label the 3D point clouds

• We advise that you label in 3D before projecting to 2D for a few reasons, even though it may seem counterintuitive. Labeling in 3D first often proves to be significantly more efficient, even if your primary interest is in obtaining 2D labels.
• Imagine you are driving past a stationary object like a traffic sign. Equipped with multiple cameras, this traffic sign remains visible in three of them as you pass by, spanning approximately 100 frames.
• Annotating this scenario using 2D bounding boxes would require you to label a total of 300 instances. However, in the 3D space, annotating this static, non-moving traffic sign would involve a single cuboid annotation.
• While labeling a 3D cuboid does take three times as long as annotating a 2D bounding box, it is still 100 times more efficient than labeling directly in the images.

Step 3: Project to 2D images

• Once you’ve labeled your 3D point clouds, you can calibrate and align them to your 2D image data. Segments.ai will automatically copy over object IDs to the 2D data, saving you hours of drawing bounding boxes. You only need to make minor adjustments during your quality control check.

La transición de un enfoque de fusión tardía a uno de fusión temprana es cada vez más común entre las empresas de vehículos autónomos y robótica en sus modelos de aprendizaje automático (machine learning, ML).

La fusión tardía implica el uso de modelos de aprendizaje automático independientes para cada sensor utilizado, que generan resultados individuales. Estos resultados se combinan o fusionan posteriormente para crear una representación 3D coherente de la escena. Consiste en ejecutar varios modelos de forma independiente y combinar sus resultados al final.

La fusión temprana adopta un enfoque más contemporáneo. En lugar de utilizar modelos separados para cada sensor, todos los datos de los sensores se integran en un solo modelo de aprendizaje automático. Este modelo unificado está diseñado para hacer predicciones directamente dentro del espacio 3D.

Para que la fusión temprana sea efectiva, la cuadrícula de vóxeles resulta ser una representación ventajosa de la escena. La cuadrícula de vóxeles presenta una estructura regular, lo que la hace adecuada para este enfoque. Se puede conceptualizar como tensores, lo que permite la predicción de extremo a extremo. Todo el proceso, desde la entrada hasta la salida, puede predecirse utilizando un solo modelo.