Task prioritized experience replay algorithm for reinforcement learning

A method of training an agent, comprising:

• performing, by the agent, an action (at) sampled from a behavior policy (πb) for an observation (st);
• storing a transition tuple in a main buffer, the transition tuple including {(st, at, {right arrow over (rt)}, st+1)}, where {right arrow over (rt)} is a reward vector for each task in an environment and st+1 is a next environment state after action (at);
• updating transiting priorities for each transition tuple stored in the main buffer;
• sampling a minibatch of transition tuples to update the task networks;
• and
• optimizing task policies from the updated task networks with an off-policy algorithm, wherein
• transitions that belong to a set of transition indices that result in achievement of task-j during an ith episode are given a priority greater than transitions that do not result in achievement of task-j during the ith episode.

A non-transitory computer-readable storage medium with an executable program stored thereon, wherein the program instructs one or more processors to perform the following steps:

• performing, by the agent, an action (at) sampled from a behavior policy (πb) for an observation (st);
• storing a transition tuple in a main buffer, the transition tuple including {(st, at, {right arrow over (rt)}, st+1)}, where t is a reward vector for each task in an environment and st+1 is a next environment state after action (at);
• updating transiting priorities for each transition tuple stored in the main buffer;
• sampling a minibatch of transition tuples to update the task networks;
• and
• optimizing task policies from the updated task networks with an off-policy algorithm, wherein
• transitions that belong to a set of transition indices that result in achievement of task-j during an ith episode are given a priority greater than transitions that do not result in achievement of task-j during the it episode.