CONTROL2016-10-05

Supervision via Competition: Robot Adversaries for Learning Tasks

Lerrel Pinto, James Davidson, Abhinav Gupta

This paper introduces a novel Robot LearningTrainingThe process of fitting a model using data or experience. approach where robots learn tasks more robustly by competing against adversarial robots, rather than through conventional supervised or self-supervised methods. For software developers, this means you can improve Robot LearningRobot learningUsing data and algorithms to help robots improve behavior instead of only relying on hand-written rules. performance by creating competitive scenarios where one Core ConceptsRobotA physical system with sensors and actuators that can observe the world and take actions. tries to disrupt another's Core ConceptsTaskThe job the robot is supposed to complete, such as pick-and-place, navigation, or drawer opening. Core ConceptsExecutionActually carrying out planned or predicted actions on the robot.—the defending Core ConceptsRobotA physical system with sensors and actuators that can observe the world and take actions. learns to be more robust as a result. The paper demonstrates this works well for Manipulation & TasksGraspingTaking hold of an object. tasks, showing 82% success on novel objects with adversarial Robot LearningTrainingThe process of fitting a model using data or experience. vs. 68% without it. This approach is particularly valuable because it solves the problem of weak supervision signals from sensors by using robot-versus-robot competition as a natural source of hard negative examples and performance pressure.

THE PROBLEM

This paper focuses on Control & PlanningControlThe method used to make the robot move the way you want.. This paper introduces a novel Robot LearningTrainingThe process of fitting a model using data or experience. approach where robots learn tasks more robustly by competing against adversarial robots, rather than through conventional supervised or self-supervised methods. For software developers, this means you can improve Robot LearningRobot learningUsing data and algorithms to help robots improve behavior instead of only relying on hand-written rules. performance by creating competitive scenarios where one Core ConceptsRobotA physical system with sensors and actuators that can observe the world and take actions. tries to disrupt another's Core ConceptsTaskThe job the robot is supposed to complete, such as pick-and-place, navigation, or drawer opening. Core ConceptsExecutionActually carrying out planned or predicted actions on the robot.—the defending Core ConceptsRobotA physical system with sensors and actuators that can observe the world and take actions. learns to be more robust as a result. The paper demonstrates this works well for Manipulation & TasksGraspingTaking hold of an object. tasks, showing 82% success on novel objects with adversarial Robot LearningTrainingThe process of fitting a model using data or experience. vs. 68% without it. This approach is particularly valuable because it solves the problem of weak supervision signals from sensors by using robot-versus-robot competition as a natural source of hard negative examples and performance pressure. Read the paper by tracking the Core ConceptsTaskThe job the robot is supposed to complete, such as pick-and-place, navigation, or drawer opening. definition, the Core ConceptsRobotA physical system with sensors and actuators that can observe the world and take actions. or data assumptions, and the evidence that supports the claimed improvement.

HOW IT WORKS

1

Task framing

The paper frames the work as Control & PlanningControlThe method used to make the robot move the way you want.. Start here because it defines what success means and which assumptions the rest of the method inherits.

2

Core method

This paper introduces a novel Robot LearningTrainingThe process of fitting a model using data or experience. approach where robots learn tasks more robustly by competing against adversarial robots, rather than through conventional supervised or self-supervised methods. For software developers, this means you can improve Robot LearningRobot learningUsing data and algorithms to help robots improve behavior instead of only relying on hand-written rules. performance by creating competitive scenarios where one Core ConceptsRobotA physical system with sensors and actuators that can observe the world and take actions. tries to disrupt another's Core ConceptsTaskThe job the robot is supposed to complete, such as pick-and-place, navigation, or drawer opening. Core ConceptsExecutionActually carrying out planned or predicted actions on the robot.—the defending Core ConceptsRobotA physical system with sensors and actuators that can observe the world and take actions. learns to be more robust as a result. The paper demonstrates this works well for Manipulation & TasksGraspingTaking hold of an object. tasks, showing 82% success on novel objects with adversarial Robot LearningTrainingThe process of fitting a model using data or experience. vs. 68% without it. This approach is particularly valuable because it solves the problem of weak supervision signals from sensors by using robot-versus-robot competition as a natural source of hard negative examples and performance pressure. When reading the method section, identify the inputs, the learned or engineered representation, and the Core ConceptsActionA command the robot sends to its motors, controller, or low-level system. or prediction produced by the system.

3

Data and supervision

For robotics work, the data story is part of the method: check whether the system depends on Imitation & Reinforcement LearningTeleoperation (teleop)A human remotely controlling the robot, often to collect demonstrations., Simulation & Sim-to-RealSimulationA virtual environment where robots can be trained or tested., internet video, human labels, or Core ConceptsRobotA physical system with sensors and actuators that can observe the world and take actions. rollouts.

4

Evaluation evidence

The paper should be judged through its Simulation & Sim-to-RealEvaluationMeasuring how well a robot system performs. protocol: what data is used, what Core ConceptsRobotA physical system with sensors and actuators that can observe the world and take actions. or simulator is tested, and which Evaluation & ResearchBaselineA reference method used for comparison. comparisons support the claim. Look for the gap between the headline result and the Simulation & Sim-to-RealDeploymentPutting the trained system on a real robot. setting you would actually care about.

KEY RESULTS

Main contributionConceptual contribution

This paper introduces a novel Robot LearningTrainingThe process of fitting a model using data or experience. approach where robots learn tasks more robustly by competing against adversarial robots, rather than through conventional supervised or self-supervised methods. For software developers, this means you can improve Robot LearningRobot learningUsing data and algorithms to help robots improve behavior instead of only relying on hand-written rules. performance by creating competitive scenarios where one Core ConceptsRobotA physical system with sensors and actuators that can observe the world and take actions. tries to disrupt another's Core ConceptsTaskThe job the robot is supposed to complete, such as pick-and-place, navigation, or drawer opening. Core ConceptsExecutionActually carrying out planned or predicted actions on the robot.—the defending Core ConceptsRobotA physical system with sensors and actuators that can observe the world and take actions. learns to be more robust as a result. The paper demonstrates this works well for Manipulation & TasksGraspingTaking hold of an object. tasks, showing 82% success on novel objects with adversarial Robot LearningTrainingThe process of fitting a model using data or experience. vs. 68% without it. This approach is particularly valuable because it solves the problem of weak supervision signals from sensors by using robot-versus-robot competition as a natural source of hard negative examples and performance pressure.

WHY DEVELOPERS SHOULD CARE

This paper introduces a novel Robot LearningTrainingThe process of fitting a model using data or experience. approach where robots learn tasks more robustly by competing against adversarial robots, rather than through conventional supervised or self-supervised methods. For software developers, this means you can improve Robot LearningRobot learningUsing data and algorithms to help robots improve behavior instead of only relying on hand-written rules. performance by creating competitive scenarios where one Core ConceptsRobotA physical system with sensors and actuators that can observe the world and take actions. tries to disrupt another's Core ConceptsTaskThe job the robot is supposed to complete, such as pick-and-place, navigation, or drawer opening. Core ConceptsExecutionActually carrying out planned or predicted actions on the robot.—the defending Core ConceptsRobotA physical system with sensors and actuators that can observe the world and take actions. learns to be more robust as a result. The paper demonstrates this works well for Manipulation & TasksGraspingTaking hold of an object. tasks, showing 82% success on novel objects with adversarial Robot LearningTrainingThe process of fitting a model using data or experience. vs. 68% without it. This approach is particularly valuable because it solves the problem of weak supervision signals from sensors by using robot-versus-robot competition as a natural source of hard negative examples and performance pressure.

LIMITATIONS

The main limitation to check is whether the claimed behavior holds outside the paper's reported setup. That means testing across different Core ConceptsRobotA physical system with sensors and actuators that can observe the world and take actions. embodiments, scenes, objects, and data distributions.

WHAT COMES NEXT

The practical next step is independent reproduction with clear baselines, ablations, and stress tests. For a developer, the useful follow-up is to map the paper's Control & PlanningControlThe method used to make the robot move the way you want. assumptions onto a concrete Core ConceptsRobotA physical system with sensors and actuators that can observe the world and take actions. stack, then test the smallest version of the method that could run end to end.

RELATED PAPERS