Robotic Manipulation Control for Occluded Object Picking

The patent introduces an intermediary computational model that bridges sensor data and robotic manipulation. This model includes virtual representations of the robot, sensors, and objects, along with a physics engine that simulates sensor data generation. The intermediary model enables the robot to predict and understand obscured objects by reasoning about what sensors would detect from different viewpoints, thereby achieving reliable manipulation even when direct sensor observation is blocked by the environment.

The patent performs preliminary computational actions by pre-building a detailed computational model that includes virtual sensors, object properties, and physics-based sensor data generation. Before actual manipulation occurs, the system pre-computes expected sensor readings from multiple viewpoints, pre-identifies objects of interest, and pre-determines their orientation and scale relative to the gripper. This preliminary processing enables fast, precise manipulation decisions during actual operation without requiring complex real-time computation.

The patent creates virtual copies of the physical system components within a computational model. Instead of using expensive advanced physical sensors, the system uses inexpensive standard sensors combined with virtual sensor models that simulate what advanced sensors would detect. The computational model includes virtual representations of objects with known properties, virtual sensors with specific detection characteristics, and a physics engine that generates synthetic sensor data. This copying approach allows the robot to achieve high measurement precision through computational reasoning rather than expensive hardware.