Michael T. Rosenstein

Approximate Optimal Control as a Model for Motor Learning

Current models of psychological development rely heavily on connectionist models that use supervised learning. These models adapt network weights when the network output does not match the target outputs computed by some agent. The authors present a model of motor learning in which the child uses exploration to discover appropriate ways of responding. The model is consistent with what is known about how neural systems evaluate behavior. The authors model the development of reaching and investigate N. Bernsteins (1967) hypotheses about early motor learning. Simulations show the course of learning as well as model the kinematics of reaching by a dynamical arm.

A FRAMEWORK FOR LEARNING AND CONTROL IN INTELLIGENT HUMANOID ROBOTS

Future application areas for humanoid robots range from the household, to agriculture, to the military, and to the exploration of space. Service applications such as these must address a changing, unstructured environment, a collaboration with human clients, and the integration of manual dexterity and mobility. Control frameworks for service-oriented humanoid robots must, therefore, accommodate many independently challenging issues including: techniques for configuring networks of sensorimotor resources; modeling tasks and constructing behavior in partially observable environments; and integrated control paradigms for mobile manipulators. Our approach advocates actively gathering salient information, modeling the environment, reasoning about solutions to new problems, and coordinating ad hoc interactions between multiple degrees of freedom to do mechanical work. Representations that encode control knowledge are a primary concern. Individual robots must exploit declarative structure for planning and must learn procedural strategies that work in recognizable contexts. We present several pieces of an overall framework in which a robot learns situated policies for control that exploit existing control knowledge and extend its scope. Several examples drawn from the research agenda at the Laboratory for Perceptual Robotics are used to illustrate the ideas.

Building an autonomous humanoid tool user

To make the transition from a technological curiosity to productive tools, humanoid robots will require key advances in many areas, including, mechanical design, sensing, embedded avionics, power, and navigation. Using the NASA Johnson Space Centers Robonaut as a testbed, the DARPA mobile autonomous robot software (MARS) humanoids team is investigating technologies that will enable humanoid robots to work effectively with humans and autonomously work with tools. A novel learning approach is being applied that enables the robot to learn both from a remote human teleoperating the robot and an adjacent human giving instruction. When the remote human performs tasks teleoperatively, the robot learns the salient sensory-motor features to executing the task. Once learned, the task may be carried out fusing the skills required to perform the task, guided by on-board sensing. The adjacent human takes advantage of previously learned skills to sequence the execution of these skills. Preliminary results from initial experiments using a drill to tighten lug nuts on a wheel are discussed.

Extracting User Intent in Mixed Initiative Teleoperator Control

User fatigue is common with robot teleoperation interfaces. Mixed initiative control approaches attempt to reduce this fatigue by allowing control responsibility to be shared between the user and an intelligent control system. A critical challenge is how the user can communicate her intentions to the control system in an intuitive manner as possible. In the context of control of a humanoid robot, we propose an interface that uses the movement currently commanded by the user to assess the intended outcome. Specifically, given the observation of the motion of the teleoperated robot for a given period of time, we would like to automatically generate an abstract explanation of that movement. Such an explanation should facilitate the execution of the same movement under the same or similar conditions in the future. How do we translate these observations of teleoperator behavior into a deep representation of the teleoperator’s intentions? Neurophysiological evidence suggests that in primates, the mechanisms for the recognition of the actions of other agents are intertwined with the mechanisms for execution of the same actions. For example, Rizzolatti et al. (1988) identified neurons within the ventral premotor cortex of monkey that fired during execution of specific grasping movements. Although this area is traditionally thought of as a motor execution area, Rizzolatti et al. (1996) showed that neurons in a subarea were active not only when the monkey executed certain grasping actions, but also when the monkey observed others making similar movements. These and other results suggest that generators of action could also facilitate the recognition of motor actions taken by another entity (in our case, the teleoperator). The foci of this study are teleoperated pick-and-place tasks using the UMass Torso robot. This robot consists of an articulated, stereo biSight head; two 7-DOF Whole Arm Manipulators (WAMs); two 3-fingered hands (each finger is equipped with a six-axis force/torque sensor); and a quadraphonic audio input system. The teleoperator interface consists of a red/blue stereo display and a P5 Essential Reality glove that senses the position and orientation of the user’s hand, as well as the flexion of the user’s fingers.

Velocity-dependent dynamic manipulability

Measures of dynamic manipulability summarize a manipulators capacity to generate accelerations for arbitrary tasks, and such measures are useful tools for the design and control of general-purpose robots. Existing measures, however, downplay the effects of velocity or else ignore them altogether. In this paper we derive the relationship between joint velocity and end-effector acceleration, and through case studies we demonstrate that velocity has a complex, non-negligible effect on manipulability. We also provide evidence that movement near a singularity is beneficial for certain tasks.

Learning at the level of synergies for a robot weightlifter

Abstract Human motor coordination is associated with “synergies” in the neuromuscular system that cause our bodies to behave as coordinated units rather than collections of decoupled degrees of freedom. The focus of this paper is the role that synergies play for robot motor coordination, and in particular, we use a weightlifting task to demonstrate how trial-and-error learning can cause synergies to evolve from a simple control system and crude kinematic path. We also show how a robot learns to exploit its intrinsic dynamics, thereby improving performance, as individual joint motions become actively coupled via the control system.

Learning what is relevant to the effects of actions for a mobile robot

We have developed a learning mechanism that allows robots to discover the conditional effects of their actions. Based on sansorimotor experience, this mechanism permits a robot to explore its environment and observe effects of its actions, These observations are used to learn a con&t operalor diflcrcncc taldc, a structure that relates circumstances (context) and actions (operators) to effects on the environmom, From the context operator difference table, one can extract a relatively small set of state variables, which simplifies tho problem of learning policies for complex activities. We demonstrate results with the Pioneer 1 mobile robot.

User intentions funneled through a human-robot interface

We describe a method for predicting user intentions as part of a human-robot interface. In particular, we show that funnels, i.e., geometric objects that partition an input space, provide a convenient means for discriminating individual objects and for clustering sets of objects for hierarchical tasks. One advantage of the proposed implementation is that a simple parametric model can be used to specify the shape of a funnel, and a straightforward heuristic for setting initial parameter values appears promising. We discuss the possibility of adapting the user interface with machine learning techniques, and we illustrate the approach with a humanoid robot performing a variation of a standard peg-insertion task.

The workshop program at the nineteenth national conference on artificial intelligence

AAAI presented the AAAI-04 workshop program on July 25-26, 2004 in San Jose, California. This program included twelve workshops covering a wide range of topics in artificial intelligence. The titles of the workshops were as follows: (1) Adaptive Text Extraction and Mining; (2) Agent Organizations: Theory and Practice; (3) Anchoring Symbols to Sensor Data; (4) Challenges in Game AI; (5) Fielding Applications of Artificial Intelligence; (6) Forming and Maintaining Coalitions in Adaptive Multiagent Systems; (7) Intelligent Agent Architectures: Combining the Strengths of Software Engineering and Cognitive Systems; (8) Learning and Planning in Markov Processes--Advances and Challenges; (9) Semantic Web Personalization; (10) Sensor Networks; (11) Spatial and Temporal Reasoning; and (12) Supervisory Control of Learning and Adaptive Systems.