Control and navigation systems, pose optimization, mapping, and localization techniques

A pose optimization process to optimize an approximate pose of a sensor with respect to an object using sensor data, the process comprising:

• receiving a model of an object;
• receiving sensor data relating to at least part of the object, the sensor data being in a sensor frame of reference, the sensor data being associated with pose data representing a pose of a sensor used to generate the sensor data;
• processing the sensor data including passing the sensor data through a convolutional neural network with a convolutional-deconvolutional architecture to generate output sensor data;
• projecting, after the processing of the sensor data, at least part of the model onto the sensor frame of reference;
• identifying corresponding feature points and/or feature lines between the model and sensor data in the sensor frame of reference, wherein identifying corresponding feature points and/or feature lines between the model and sensor data in the sensor frame of reference includes identifying corresponding feature points and/or feature lines using the model and the output sensor data;
• optimizing the pose data based on the identified correspondence, using a location difference in the corresponding feature points and/or feature lines between the model and the sensor data in the sensor frame of reference;
• and
• optimizing at least one parameter of the model of the object, wherein the at least one parameter defines an estimated configuration of the object, and wherein optimizing the at least one parameter includes optimizing the at least one parameter based on the correspondence between the feature points and/or feature lines and the optimized pose data, such that the optimized parameter represents the configuration of the object.

A navigation process for an autonomous vehicle, the navigation process including:

• receiving a model of an object;
• receiving sensor data relating to at least part of the object, the sensor data being in a sensor frame of reference, the sensor data being associated with pose data representing a pose of a sensor used to generate the sensor data;
• projecting at least part of the model onto the sensor frame of reference;
• identifying corresponding feature points and/or feature lines between the model and sensor data in the sensor frame of reference, wherein identifying corresponding feature points and/or feature lines between the model and sensor data in the sensor frame of reference includes defining one or more line points along a feature line identified in the model, defining search line and direction, and searching in accordance with the search line and search direction for one or more respective corresponding line points;
• optimizing the pose data based on the identified correspondence, using a location difference in the corresponding feature points and/or feature lines between the model and the sensor data in the sensor frame of reference;
• and
• defining or modifying a navigation plan using the optimized pose data.