Hi,
I am using the Learning Agents plugin and have extended the learning to drive tutorial into C++ and to add obstacle avoidance.
The simulation begins as expected with clear progress made, but after 5-10 minutes something falls out of sync and I get thousands of log warnings and errors. I have generalized the log output below, but I get one line for every completion/ reward/ observation/ action once the iteration desyncs.
LogLearning: Error: Agent Trainer: Completion “CompletionVar” for agent with id “n” has not been set (got iteration “N”, expected iteration “N+1”) and so agent will not have completions evaluated.
LogLearning: Error: Agent Trainer: Reward “RewardVar” for agent with id “n” has not been set (got iteration “N”, expected iteration “N+1”) and so agent will not have rewards evaluated.
LogLearning: Warning: Agent Trainer: Agent with id “n” has non-matching iteration numbers (observation: “n”, action: “n”, reward: “n-1”, completion: “n-1”). Experience will not be processed for it.
LogLearning: Warning: Agent Interactor: Observation “ObservationVar” for agent with id “n” has not been set (got iteration “N”, expected iteration “N+1”) and so agent will not have observations encoded.
LogLearning: Warning: Agent Interactor: Action “ActionVar” for agent with id “n” appears to have not been got.
I guess my question is how should I debug this, and is there a likely cause of this error?