We are building a humanoid robot at the MIT Artificial Intelligence Laboratory. It is a legless human sized robot and is meant to be able to emulate human functionality. We are particularly interested in using it as vehicle to understand how humans work, by trying to combine many theories from artificial intelligence, cognitive science, physiology and neuroscience. While undergoing continual revisions in its hardware and software the robot has been operational in one form or another for over three years. We are working on systems that emulate both human development and human social interactions.

View full abstract

Desyncronized tele-robotics by virtual reality technology has been studied by many researchers. This approach does not couple an operator and a slave robot directly like bilateral master-slave teleoperation system, so the smart search of the remote environment by autonomous slave robots is strongly required. In this paper, we propose a method which can measure the mechanical impedance and shape of a target object and environment with the robot under impedance control by adding perturbation to the virtual trajectory. First, it is considered what kind of perturbation is necessary and realistic for that purpose. Then, we show that the robot with a single force sensor is insufficient for simultaneous estimation of mechanical impedance and shape, and that the measurement of force at plural points on the contact surface is essenstial. The validity of the method is checked by a 2DOF real robot.

View full abstract

This research is concerned with a compensation method of a mobile manipulator using a neural network for moving operation and with the tracking control performance. The mobile manipulator can spread its working area widely and improve its working efficiency. Several papers of mobile manipulators have been presented and assumed that the traveling terrain is flat. But, irregularities exist even on an artificial floor in factories and buildings, and it is difficult to measure exactly the shape of the terrain. When the end-effector of the mobile manipulator tracks desired trajectory during the mobile manipulator travels on an unknown irregular terrain, it is subjected to the effect of the disturbance torques derived from influences depending on the terrain. Furthermore, the disturbance torque derived from the change of payload disturbs the accurate trajectory control. Then, a compensation method for the two kinds of the disturbance torque must be considered. First, for the change of the payload, the adaptive controller is constructed for the mobile manipulator. Second, the control performance of the adaptive controller is examined. Third, for the disturbance depending on the terrain, the compensator using a neural network is proposed. Finally, the validity of the adaptive controller with a neuro compensator is clarified through simulation experiments.

View full abstract

Manipulators are equipped with several joints and it is possible to operate along any given trajectories. This flexibility is essential advantages and valuable studies are conducted for trajectory planning to effectively control manipulators. However, manipulators belong to a class of complex nonlinear dynamic systems and the application of trajectory planning leads to substantial difficulties. Therefore, various methods are proposed to solve optimal problems by adding or neglecting adequate constraints. However, they do not seem to offer generalized trajectory formation methods.
This paper discusses trajectory planning of point-to-point operations under condition that only an initial and a final states are given. Performance criterion is defined by consumed energy dissipated in operations and motor torque limit is considered. Mathematical formulation to calculate optimal trajectories is proposed and simulation is conducted for a 2 degrees of freedom manipulator. Amount of energy dissipation is discussed in relation with optimal and various operating time for typical manipulator movements. Besides, merit of trajectory planning based on a minimum-energy criterion and operating time is stated. Finally, characteristics of energy consumption in human arm movements is discussed and formulated trajectories are verified that they show similar features with human arm trajectories in relatively fast movements.

View full abstract

Hyper-redundant manipulator has a very large or infinite degrees of kinematic redundancy, thus possesses some unconventional features such as the ability to enter a narrow space while avoiding obstacles. In this paper, we propose a new technique for a planar hyper-redundant manipulator to avoid the existing static obstacles in environment while performing a payload-location task from point to point. It is based on analysis in the defined posture space, where three parameters were used to determine the hyper-redundant manipulator configuration. The scheme is verified by computer simulation in case of using the model of the developed Hyper-R Arm. It shows that our method works perfect and the obstacles are well avoided globally.

View full abstract

A model-based system to implement force-feedback based teleoperation with long time delay is presented. Such a system usually requires a precise model of the real world. However, the precise model is difficult to be realized; the system should be designed with some robustness against errors. In this paper, we propose a model-based space teleoperation system with robustness against geometrical modeling errors. The system consists of a master arm, a slave arm, and a virtual arm which are controlled independently by velocity command. Each arm has autonomy to switch between force and velocity control for contact and non-contact modes, respectively. The operator can feel contact force reflected from the virtual arm. The slave arm realizes the commanded force and, consequently, safe operation is guaranteed. By using a master arm without backdrivability and with high gear ratio, the operator can feel hard contact very well. The experimental results show that a sample ORU task and a door opening task can be realized very effectively, in spite of the long time delay.

View full abstract

To capture a target on orbit is a fundamental task for space robots. As the target doesn't have a fixed base, it might be bounced away before the capture procedure is completed. When the target is much heavier than the robot hand, the manipulator controller gain affects such dynamics. This paper uses two measures for evaluation, namely, 1) contact time (duration until the target is bounced off) and 2) final relative velocities (target velocities relative to the hand when the target drifts away) . Analytical approximations of the two measures are derived and validated through numerical simulations and hardware experiments.

View full abstract

A new type of cooperation called mutual handling is introduced in this paper, which enables a multi-robot system to achieve an advanced functionality that cannot be realized by a single robot, and as an application of mutual handling, a method of cooperative transportation for two autonomous mobile robots to climb over a large step by operating each forklifts cooperatively is proposed, which cannot be overcome by a single robot. Then, a forklift mechanism for mutual handling is designed, and two autonomous mobile robots equipped with the forklifts are developed. Finally, a cooperative transportation method in which two autonomous mobile robots are controlled based on communication is presented, and the step-climbing motion by cooperative transportation is proved feasible by showing the experimental results using the developed omni-directional mobile robots.

View full abstract

In this paper, we propose an environment for developing autonomous robot brains in which the developer can describe a brain based on an abstract model in a generic programming language, and verify it immediately on a multi-processor system. We define BeNet, an abstract module network model, for describing a robot brain completely and claim that it is an appropriate model for designing brains and simulating them on multi-processor systems. Since a BeNet is a static network of asynchronous modules that interact each other at regular intervals, it represents a dynamical system completely and concisely and can be simulated efficiently by various multi-processor systems. You can describe a brain frexibly as a BeNet out of desirable modules. We describe our model and show how to implement our environments for developing BeNets for robot brains to show that they are simulated efficiently by multi-processor systems. Our environments make it possible to describe brains in C or C++ in a simplified manner, verify the behaviors of them in the real world and modify them in a short term to develop desirable systems.

View full abstract

The metal hydride (MH) actuator is based on a new concept using the reversible reaction between the heat energy and mechanical energy of a hydrogen absorbing alloy. As the MH actuator is driven by heat, the response speed depends on the heat conductivity of the alloy layer in the MH container. Changing the alloy layer thickness directly influences the hydrogen absorption speed, and the addition of Cu-powder to the alloy layer contributes to improve heat conductivity.
The MH actuator is characterized by large output per weight of the unit, that is, compactness, noiseless operation and smooth action. As an actuator suitable for use in rehabilitation equiment, we have developed a large-output MH actuator that is driven by heat from eletric heaters and tapwater, and then applied it to a toilet seat lifter and investigated its adaptability.
It is expected that such MH actuators will play important roles in the development of various types of rehabilitation equipment.

View full abstract

A new method to generate familiar behavior of a robot is proposed and verified. Familiar behavior of a robot is important under human-robot symbiosis in order to make a user feel comfortable and to avoid putting stress on a user. The method described is based upon a hypothesis that passive reactions produce familiar and comfortable impression in human beings. The results of psychological experiments verifies the validity of the method. A stuffed robot is made for the experiments. Impressions produced by the behaviors of the robot are analyzed quantitatively using the semantic differential method. Utilizing the robot behavior for generation familiarity will facilitate human-robot symbiosis.

View full abstract

We propose a position estimation method for non-stop navigation of an autonomous mobile robot. The proposed method is based on the maximum likelihood estimation. To cope with the time delay in the sensor data process, we introduce a retroactive positioning data fusion method. The proposed technique is implemented on our small size autonomous mobile robot. An experimental result is shown, in which the robot could navigate itself without stopping even when it takes several seconds of processing time to recognize landmark from external sensor data.

View full abstract

With the depletion of global energy resources, saving energy has become an important aspect in design of energy consuming equipment. Hence, this paper deals with an optimal time of operation of a manipulator which minimizes the energy dissipated in the actuators. When the manipulator is operated in a vertical plane, the system is highly non-linear due to gravity and an analytical solution can not be found. Therefore, the system is first linearized around various equilibrium points and the existence of an optimal time is investigated theoretically. The theoretical results resemble the simulation of the non-linear system except for a slight difference in the optimal time. The theoretical results and simulations show that an optimal operating time exists when the first link of the manipulator traverses the stable equilibrium point and that it does not exist when the first link traverses the unstable equilibrium point. Finally, a non-linear optimization method is applied to ascertain the accurate optimal time of operation.

View full abstract

This paper describes one of the control methods which attain robust positioning performances of a parallel manipulator. Pneumatic cylinders are employed as a driving actuator by focusing on its rectilinear movement which makes it easy to construct a manipulator and high power/weight ratio and so on. In positioning control schemes of a parallel manipulator, almost of them are based on the inverse kinematics, a map from a position/orientation of the platform to displacement of links, since inverse kinematics has an analytic solution contrary to the serial link manipulator. This method has an advantage that a complicated manipulator's dynamics is not needed, but the influences of interaction force and manipulator's dynamics, which act as disturbances on the positioning servo, must be lowered to improve control performances. Basically pneumatic actuator is controlled by pressure, therefore if pressure would be generated so as to cancel the influence of disturbances, then disturbances would not affect on the control performances entirely. Therefore positioning control system employing 2 disturbance observers is constructed based on the pressure control. If dynamics of the platform could be adjusted based on the hand coordinate system, its field of use would be spread out extremely. In this study, a control system based on the hand coordinate system is also constructed. The validity of these control systems are confirmed through some experiments and analysis.

View full abstract