Autonomous vehicles are being utilized across numerous industries resulting in dramatic increases in business performance. To operate safely and successfully, these autonomous vehicles must navigate their surroundings without incident. As such, collision avoidance is one of the largest areas of concern when it comes to the implementation of autonomous vehicles.
Accurate and reliable detection of obstacles is paramount to collision avoidance, but can be quite challenging as obstructions tend to vary in size and geometry. Some objects, such as dust, can cause a false-positive or false-negative reaction, when in reality, there is no impending collision.
The problem of obstacle detection can be reduced to a two-state Markov model, which is a model used for randomly changing systems based on stochastic calculus. The transition between these two states, obstacle or non-obstacle, is then governed by a Bayesian inference which takes into account past probability, current measurements, and sensor characteristics. Thus, the most comprehensive way to correctly understand the real obstacles autonomous vehicles encounter is by applying sensor fusion.
By “fusing” LiDAR and Radar measurements into a complete occupancy grid, along with advanced Artificial Intelligence (AI) and Machine Learning (ML), ASI’s Obstacle Detection Sensor Fusion application optimizes obstacle detection for increased collision avoidance. Through the implementation of predictive analytics which account for prior inputs of cell occupancy probability, a labeled 3D point-cloud from the LiDAR unit, the detected objects by the radar, and the associated radar field of view, ASI’s algorithm builds a complete grid system that identifies, classifies, and circumvents obstacles.
The value in each grid cell is computed using the measurements that are classified as either obstacle or non-obstacle. Then, based on adjustments to both true-positive and true-negative rates, the AI algorithm is tuned for each sensor by using the associated confidence level. This method has been show to enhance collision avoidance.
ASI’s Obstacle Detection Sensor Fusion enables autonomous vehicles to traverse unknown terrain and dangerous obstacles to arrive at a desired endpoint safely. The advanced AI and ML algorithm creates an accurate model of the world, identifies and classifies obstacles and non-obstacles, then stores this information in a database which can be accessed by any autonomous vehicle that enters that region, thereby allowing for continuous updating and learning. Teaching a vehicle to learn is just the beginning — ASI’s research and development team is taking this technology further by enabling autonomous vehicles to share information for a collaborative improvement in vehicle behavior.
Learn more about ASI’s collision detection at work in the mining industry and current areas of research.