Precise robotic manipulation skills are desirable in many industrial settings, reinforcement learning (RL) methods hold the promise of acquiring these skills autonomously. In this paper, we explicitly consider incorporating operational space force/torque information into reinforcement learning; this is motivated by humans heuristically mapping perceived forces to control actions, which results in completing high-precision tasks in a fairly easy manner. Our approach combines RL with force/torque information by incorporating a proper operational space force controller; where we also exploit different ablations on processing this information. Moreover, we propose a neural network architecture that generalizes to reasonable variations of the environment. We evaluate our method on the open-source Siemens Robot Learning Challenge, which requires precise and delicate force-controlled behavior to assemble a tight-fit gear wheel set.

This work introduces a novel 12-motor paired-cable actuation scheme to achieve rolling locomotion with a spherical tensegrity structure. Using a new point mass tensegrity dynamic formulation which we present, we utilize Model Predictive Control to generate optimal state-action trajectories for benchmark evaluation. In particular, locomotive performance is assessed based on the practical criteria of rolling speed, energy efficiency, and directional trajectory-tracking accuracy. Through simulation of 6-motor, 12-motor paired cable, and 24-motor fully-actuated policies, we demonstrate that the 12-motor schema is superior to the 6-motor policy in all benchmark categories, comparable to the 24-motor policy in rolling speed, and is over five times more energy efficient than the fully-actuated 24-motor configuration.

Tensegrity robots are composed of rigid rods connected by elastic cables, and their unique light-weight yet compliant structure makes them an appealing choice for space exploration. However, locomotion control for these robotic systems remains difficult due to their nonlinear dynamics and high-dimensional state space. We demonstrate that in the domain of tensegrity robotics, it is possible to efficiently learn end-to-end locomotion policies using mirror descent guided policy search (MDGPS) even with limited sensory inputs. We compare learned neural network policies with other locomotion control policies in various testing environments; and results show that neural network policies consistently outperform others. We also shed light to the policy learning process by analyzing different choices of observation inputs to the robot. Moreover these findings motivate exploration of deep reinforcement learning algorithms in the domain of tensegrity robotics. We show preliminary results with one such locomotion example on discontinuous rough terrains.

This work presents a model-based approach for creating robust control policies for rolling locomotion with a spherical tensegrity topology. Utilizing the structured dynamics of Class-1 tensegrity systems, we turn to model predictive control (MPC) to generate optimal multi-cable actuation trajectories for dynamic rolling. Although the resulting multi-cable state-action trajectories successfully outperform the benchmark single-cable policy performance in speed, computational constraints prevent MPC from being applied in real-time. To address this, we demonstrate that a contextual policy trained using supervised deep learning on the generated optimal MPC trajectories can be used as an end-to-end feedback policy for real-time directed rolling locomotion.

Reinforcement learning for assembly tasks can yield powerful robot control algorithms for applications that are challenging or even impossible for “conventional” feedback control methods. Insertion of a rigid peg into a deformable hole of smaller diameter is such a task. In this contribution we solve this task with Deep Reinforcement Learning. Force-torque measurements from a robot arm wrist sensor are thereby incorporated two-fold; they are integrated into the policy learning process and they are exploited in an admittance controller that is coupled to the neural network. This enables robot learning of contact-rich assembly tasks without explicit joint torque control or passive mechanical compliance. We demonstrate our approach in experiments with an industrial robot.

This paper presents a new teleoperated spherical tensegrity robot capable of performing locomotion on steep inclined surfaces. With a novel control scheme centered around the simultaneous actuation of multiple cables, the robot demonstrates robust climbing on inclined surfaces in hardware experiments and speeds significantly faster than previous spherical tensegrity models. This robot is an improvement over other iterations in the TT-series and the first tensegrity to achieve reliable locomotion on inclined surfaces of up to 24°. We analyze locomotion in simulation and hardware under single and multi-cable actuation, and introduce two novel multi-cable actuation policies, suited for steep incline climbing and speed, respectively. We propose compelling justifications for the increased dynamic ability of the robot and motivate development of optimization algorithms able to take advantage of the robot's increased control authority.

This work presents a novel spherical tensegrity robot, T12-R, which is designed and prototyped based on a twelve-rod tensegrity structure that resembles a rhombicuboctahedron. The geometry of T12-R allows for fast rolling and a detailed description of the robot design is provided. A simulation study of T12-R shows that the robot is capable of performing static locomotion. Control strategies for achieving dynamic locomotion in hardware are discussed.

Maximising electricity production from wind requires improvement of wind turbine reliability. Component failures result in unscheduled or reactive maintenance on turbines which incurs significant downtime and, in turn, increases production cost, ultimately limiting the competitiveness of renewable energy. Thus, a critical task is the early detection of faults. To this end, we present a framework for fault detection using machine learning that uses Supervisory Control and Data Acquisition (SCADA) data from a large 3MW turbine, supplemented with features derived from this data that encapsulate expert knowledge about wind turbines. These new features are created using application domain knowledge that is general to large horizontal-axis wind turbines, including knowledge of the physical quantities measured by sensors, the approximate locations of the sensors, the time series behaviour of the system, and some statistics related to the interpretation of sensor measurements. We then use mRMR feature selection to select the most important of these features. The new feature set is used to train a support vector machine to detect faults. The classification performance using the new feature set is compared to performance using the original feature set. Use of the new feature set achieves an F1-score of 90%, an improvement of 27% compared to the original feature set.

This work presents a 10 kg tensegrity ball probe that can quickly and precisely deliver a 1 kg payload over a 1 km distance on the Moon by combining cable-driven rolling and thruster-based hopping. Previous research has shown that cable-driven rolling is effective for precise positioning, even in rough terrain. However, traveling large distances using thruster-based hopping, which is made feasible by the lightweight and compliant nature of the tensegrity structure, has not been explored. To evaluate the feasibility of a thruster-based tensegrity robot, a centrally-positioned cold gas thruster with nitrogen propellant was selected, and the system was simulated using the NASA Tensegrity Robotics Toolkit (NTRT) for four hopping profiles on hilly terrains. Optimizing energy efficiency and mechanical capabilities of the tensegrity robot, hopping profiles with a long flight distance per hop, followed by the higher accuracy rolling, are recommended. Simulations also show that thrust regulation can improve energy efficiency. Regulation of thrust magnitude can be achieved using a pressure regulator, but regulation of thrust orientation calls for additional control effort. In this paper, it is demonstrated that gimbal systems as well as shape-shifting control of the tensegrity structure have the potential to regulate thrust orientation. Finally, algorithms for localization and path planning that combine hopping and rolling for energy-efficient navigation are presented.

The Spherical Underactuated Planetary Exploration Robot ball (SUPERball) is an ongoing project within NASA Ames Research Center's Intelligent Robotics Group and the Dynamic Tensegrity Robotics Lab (DTRL). The current SUPERball is the first full prototype of this tensegrity robot platform, eventually destined for space exploration missions. This work, building on prior published discussions of individual components, presents the fully-constructed robot. Various design improvements are discussed, as well as testing results of the sensors and actuators that illustrate system performance. Basic low-level motor position controls are implemented and validated against sensor data, which show SUPERball to be uniquely suited for highly dynamic state trajectory tracking. Finally, SUPERball is shown in a simple example of locomotion. This implementation of a basic motion primitive shows SUPERball in untethered control.

Robots based on tensegrity structures have the potential to be robust, efficient and adaptable. While traditionally being difficult to control, recent control strategies for ball-shaped tensegrity robots have successfully enabled punctuated rolling, hill-climbing and obstacle climbing. These gains have been made possible through the use of machine learning and physics simulations that allow controls to be “learned” instead of being engineered in a top-down fashion. While effective in simulation, these emergent methods unfortunately give little insight into how to generalize the learned control strategies and evaluate their robustness. These robustness issues are especially important when applied to physical robots as there exists errors with respect to the simulation, which may prevent the physical robot from actually rolling. This paper describes how the robustness can be addressed in three ways: 1) We present a dynamic relaxation technique that describes the shape of a tensegrity structure given the forces on its cables; 2) We then show how control of a tensegrity robot “ball” for locomotion can be decomposed into finding its shape and then determining the position of the center of mass relative to the supporting polygon for this new shape; 3) Using a multi-step Monte Carlo based learning algorithm, we determine the structural geometry that pushes the center of mass out of the supporting polygon to provide the most robust basic mobility step that can lead to rolling. Combined, these elements will give greater insight into the control process, provide an alternative to the existing physics simulations and offer a greater degree of robustness to bridge the gap between simulation and hardware.

Human-centered design teams must integrate diverse individual perspectives into a shared understanding during conceptual design. The team's shared knowledge of their users becomes the basis for later design decisions. We conducted a formative study that shows how generic groupware is insufficient to support the transition from individual to collaborative creative work. We developed a set of design guidelines and implemented them in Dazzle, a collaborative shared display system for co-located design team meetings. Dazzle associates the action of showing information on the shared display with granting the rest of the team access to that information: showing is sharing. Dazzle also records a history of shown files. Team members can annotate this log using cross-platform synchronized clients. Teams of novice designers tested Dazzle over two consecutive sessions: the first focused on synthesizing user research, and the second focused on brainstorming. Dazzle was very effective at grounding team conversations about user research, but was used less for sharing information during brainstorming tasks. Items from the shared activity log were used as sources of inspiration and decision criteria during the brainstorming task. Future work includes additional support for active decision-making, and ambient feedback on design activity.

We present GreenHat, an interactive mobile learning application that helps students learn about biodiversity and sustainability issues in their surroundings from experts' points of view, before participating in unfamiliar debates about their familiar surroundings. Using the interactive location-sensitive map and video on a smart phone, GreenHat simulates how experts go about making observations in the field and encourages students to actively observe their environment. We present our design process, our initial prototype, report the results from our preliminary evaluation, and discuss ongoing work.

Are you trying to integrate interactive simulations, applets, case studies, courseware or other web-accessible materials into your classes? Where do you go to find these digital learning materials? How do you evaluate the quality of the materials you do find? Are there digital learning materials available that are aligned with the ABET criteria? Are there related resources for assessing student outcome assessments that you can use? How can you customize your course website with supplemental materials for students? How can you find a collection of self-studies that can be used to guide a department as they prepare for the ABET review process? This workshop introduces faculty who are interested in integrating digital learning materials in their courses to a set of criteria and methods useful in selecting and evaluating the quality of these materials to help achieve their course goals. The workshop focuses on the 10,000 educational resources cataloged in the Engineering Pathway digital library (www.engineeringpathway.org) and goes through the resources and tools available for faculty to use to locate, evaluate and select helpful digital learning materials to achieve their teaching and learning goals. Participants will be introduced to a general intellectual framework for integrating digital learning materials that stresses identifying the particular learning objectives and pedagogies for the use of particular materials. Participants will be introduced to two sets of evaluation criteria, those used in the Premier Award for Excellence in Engineering Education Courseware and another set that is used to guide catalogers as they register materials in a digital library. They will have a hands-on opportunity to apply these criteria to better understand the metrics for quality in digital learning materials, and how to apply these metrics to materials they are considering using to help achieve their course goals. The workshop also helps faculty locate courseware that can help satisfy the ABET criteria for evaluation. For those preparing for the ABET review process, the Engineering Pathway identifies a number of self-study that can be used to guide departments in the development of measurement instruments and processes. Lastly, the workshop will introduce new tools for student engagement with history of technology and well as current news in each discipline. Annotated textbooks with links to context-sensitive links to Engineering Pathway resources will also be explored.

Multiobjective genetic algorithms (MOGA) and case-based reasoning (CBR) have proven successful in the design of MEMS (microelectromechanical systems) suspension systems. Object-oriented data structures of primitive and complex genetic algorithm (GA) elements have been developed to restrict genetic operations to produce feasible design combinations as required by physical limitations or practical constraints. Thus, virtual linkage between genes and chromosomes are coded into the properties of pre-defined GA objects. A new design problem requires selecting the right primitive elements, associated data structures, and linkages that promise to produce the best gene pool for new functional requirements. In this paper, biomimetics is proposed as a means to examine and classify functional requirements so that case-based reasoning algorithms can be used to map design requirements to promising initial conceptual designs and appropriate GA primitives. The concept is demonstrated using micro-mechanical resonators.

This paper presents the use of Interactive Evolutionary Computation (IEC) as a method to allow a human user to embed their expert domain knowledge and experience to overcome the deficiencies of Modified Nodal Analysis (MNA)-based EC in a fraction of the time associated with computationally expensive Finite Element Modeler (FEM)-based EC. From our comparison tests for the design of a MEMS resonating mass, results show that MNA-based IEC performance was within 10% of FEM-based EC, yet it performed in only 1/24thof the computation time.

We describe two approaches to reducing human fatigue in interactive evolutionary computation (IEC). A predictor function is used to estimate the human user's score, thus reducing the amount of effort required by the human user during the evolution process. The fuzzy system and four machine learning classifier algorithms are presented. Their performance in a real-world application, the IEC-based design of a micromachine resonating mass, is evaluated. The fuzzy system was composed of four simple rules, but was able to accurately predict the user's score 77% of the time on average. This is equivalent to a 51 % reduction of human effort compared to using IEC without the predictor. The four machine learning approaches tested were k-nearest neighbors, decision tree, AdaBoosted decision tree, and support vector machines. These approaches achieved good accuracy on validation tests, but because of the great diversity in user scoring behavior, were unable to achieve equivalent results on the user test data.

An automated design synthesis system based on a multi-objective genetic algorithm (MOGA) has been developed for the optimization of surface-micromachined MEMS devices. A hierarchical component-based genotype representation is used, which incorporates specific engineering knowledge into the design and optimization process. Each MEMS component is represented by a gene with its own parameters defining its geometry and the way it can be modified from one generation to the next. The object-oriented genotype structures efficiently describe the hierarchical nature typical of engineering designs. They also prevent MOGA from wasting time exploring inappropriate regions of the search space. The automated MEMS design synthesis is demonstrated with surface-micromachined resonator and accelerometer designs. (Track Category: EVOLUTIONARY MULTIOBJECTIVE OPTIMIZATION)

We propose an interactive layout synthesis CAD tool. It allows a human user to graphically manipulate and simulate a layout, but also has the ability to apply two simple evolutionary synthesis methods: a simple random walk function and simulated annealing. These methods are capable of optimizing very complex functions difficult for a human to tune by hand. Additionally the tool incorporates the ability to reduce the dimension of the search space via variable locking, contributing to both focused search and faster convergence to a desired performance. The CAD tool has been written with flexibility in mind, allowing the application to a wide range of layout synthesis problems. In this paper its effectiveness is demonstrated on a MEMS vibratory rate gyroscope example.

The focus of this work is on the design of a system for informal education in rural, farmworker populations using mobile devices. We have conducted needs assessment with farmworkers in the California Central Valley in conjunction with engineering and industrial design students as part of a service learning initiative. The community, working with the students, identified key needs related to accessing information, emphasizing health and legal rights. We propose the possibility of wireless access to digital libraries for access to this information. We discuss an appropriate system design using mobile phones and future plans for user testing with the community. We stress the importance of continually working with the community to develop relevant and sustainable solutions.

A test-feedback strategy is described for improving evolutionary synthesis based on the results from the fabrication and characterization of output from a genetic algorithm(GA)-based synthesis program. Simulator and fabrication conditions lead to certain configurations of output differing significantly from predicted performance when fabricated and measured. Using lessons learned from a microelectromechanical system (MEMS) synthesis characterization study, four modifications to the objectives and constraint settings of the GA formulation are evaluated to produce results that more closely match desired performance when fabricated. Statistical tests show improvement in the quality of the GA's output, producing designs with significantly less simulation and fabrication errors using the proposed methods.

This paper introduces fuzzy belief nets (FBN). The ability to invert arcs between nodes is key to solving belief nets. The inversion is accomplished by defining closeness measures which allow diagnostic reasoning from observed symptoms to cause of failures. The closeness measures are motivated by a Lukasiewicz operator which takes into account the distance from an observed symptom set to the modeled symptom set for all failure combinations. Hypothesized failures are then ranked according to maximum closeness measure and minimum cover, i.e., number of faults. Within the realm of fuzzy logic we show the graphical representation and solution of fuzzy belief nets.

The Synthesis Engineering Education Coalition strives to integrate multidisciplinary, open-ended problem solving into the varied engineering curricula of its members. To achieve this goal, Synthesis has developed a broad array of computer-based multimedia courseware modules and elements. In addition to the courseware, Synthesis has developed NEEDS, the National Engineering Education Delivery System, as the infrastructure for disseminating these and other education materials over the Internet. Several interesting challenges have been identified through this effort: How can electronic courseware meet the diverse needs of curricula among a cross section of universities? How do educators adapt traditional teaching roles to fit new resources and delivery styles? What courseware access modes equally suit the needs of author, teacher, and student? Can an infrastructure designed for static course-ware be adapted to dynamically changing information on the World Wide Web? The experience of Synthesis/NEEDS can begin to answer these questions while opening more issues in distance independent education. © 1996 John Wiley & Sons, Inc.

An interface for software that creates a natural environment for engineering graphics students to improve their spatial reasoning and 3D visualization skills is described. The skills of interest involve spatial transformations and rotations, specifically those skills that engineers use to reason about 3D objects based on 2D representations. The software uses an intuitive and interactive interface allowing direct manipulation of objects. Animation capability is provided to demonstrate the relationship between arbitrary positions of an object and standard orthographic views. A second skill of interest requires visualization of a cutting-plane intersection of an object. An interface is developed which allows intuitive positioning of the cutting-plane utilizing the metaphor of a “pool of water” in which the object is partially submerged. The surface of the water represents the cutting plane. Adjustment of the pool depth combined with direct manipulation of the object provides for arbitrary positioning of the cutting-plane. Subjective evaluation of the software thus far indicates that students enjoy using it and find it helpful. A formal testing plan to objectively evaluate the software and interface design is underway.

The authors present the concept of influence diagrams for representing probabilistic dependence and independence between state variables in a given problem domain and a topological framework for solving probabilistic inference problems in expert systems. The mathematical basis for influence diagrams is explained and theorems for mathematical manipulation of them are presented, in a graph-theoretic framework. Topological transformation rules developed in previous research are formalized in an axiomatic manner based on a concept of consistency. A polynomial-time symbolic-level algorithm for solving probabilistic inference problems is developed. The algorithm involves searching through the diagram to answer any specific diagnostic query about the system.>