Techniques for controlling robots using dynamic gain tuning

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:

• inputting a target return and a previous robot trajectory into a first trained machine learning model that outputs a robot motion and a predicted force associated with the robot motion, wherein the previous robot trajectory comprises previous robot motions and forces sensed by a force sensor;
• inputting the robot motion, the predicted force, and the previous robot trajectory into a second trained machine learning model that outputs an admittance gain associated with the predicted force;
• generating one or more robot commands based on the robot motion and the admittance gain;
• and
• causing a robot to move based on the one or more robot commands.

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:
• input a target return and a previous robot trajectory into a first trained machine learning model that outputs a robot motion and a predicted force associated with the robot motion, wherein the previous robot trajectory comprises previous robot motions and forces sensed by a force sensor,
• input the robot motion, the predicted force, and the previous robot trajectory into a second machine learning model that outputs an admittance gain associated with the predicted force,
• generate one or more robot commands based on the robot motion and the admittance gain, and
• cause a robot to move based on the one or more robot commands.