Simulation and reinforcement learning techniques for performing robotic tasks

One or more non-transitory computer-readable media storing instructions that, when executed by at least one processor, cause the at least one processor to perform the steps of:

• performing a plurality of iterations of reinforcement learning to train an untrained machine learning model to generate a trained machine learning model, wherein performing each iteration included in the plurality of iterations comprises:
• performing a plurality of simulations of a robot interacting with one or more objects represented by one or more signed distance functions (SDFs), wherein performing the plurality of simulations comprises reducing a number of contacts between the one or more objects that are being simulated by:
• assigning one or more candidate contacts included in a set of candidate contacts between the one or more objects to a set of existing contact patches based on surface normals of the set of candidate contacts,
• identifying a first candidate contact having a deepest penetration associated with the one or more objects among remaining candidate contacts that have not been assigned to any existing contact patch in the set of existing contact patches, and
• creating a new contact patch to add to the set of existing contact patches based on the first candidate contact by assigning, to the new contact patch, one or more candidate contacts included in the remaining candidate contacts having surface normals similar to the first candidate contact;
• and
• updating one or more parameters of the untrained machine learning model based on the plurality of simulations.

A system, comprising:

• one or more memories storing instructions;
• and
• one or more processors that are coupled to the one or more memories and, when executing the instructions, are configured to:
• perform a plurality of iterations of reinforcement learning to train an untrained machine learning model to generate a trained machine learning model, wherein performing each iteration included in the plurality of iterations comprises:
• performing a plurality of simulations of a robot interacting with one or more objects represented by one or more signed distance functions (SDFs), wherein performing the plurality of simulations comprises reducing a number of contacts between the one or more objects that are being simulated by:
• assigning one or more candidate contacts included in a set of candidate contacts between the one or more objects to a set of existing contact patches based on surface normals of the set of candidate contacts,
• identifying a first candidate contact having a deepest penetration associated with the one or more objects among remaining candidate contacts that have not been assigned to any existing contact patch in the set of existing contact patches, and
• creating a new contact patch to add to the set of existing contact patches based on the first candidate contact by assigning, to the new contact patch, one or more candidate contacts included in the remaining candidate contacts having surface normals similar to the first candidate contact;
• and
• updating one or more parameters of the untrained machine learning model based on the plurality of simulations.