As general purpose robots become more capable, pre-programming of all tasks at the factory will become less practical. We would like for non-technical human owners to be able to communicate, through interaction with their robot, the details of a new task; I call this interaction "task communication". During task communication the robot must infer the details of the task from unstructured human signals, and it must choose actions that facilitate this inference. In this dissertation I propose the use of a partially observable Markov decision process (POMDP) for representing the task communication problem; with the unobservable task details and unobservable intentions of the human teacher captured in the state, with all signals from the human represented as observations, and with the cost function chosen to penalize uncertainty. This dissertation presents the framework, works through an example of framing task communication as a POMDP, and presents results from a user experiment where subjects communicated a task to a POMDP-controlled virtual robot and to a human controlled virtual robot. The task communicated in the experiment consisted of a single object movement and the communication in the experiment was limited to binary approval signals from the teacher. This dissertation makes three contributions: 1) It frames human-robot task communication as a POMDP, a widely used framework. This enables the leveraging of techniques developed for other problems framed as a POMDP. 2) It provides an example of framing a task communication problem as a POMDP. 3) It validates the framework through results from a user experiment. The results suggest that the proposed POMDP framework produces robots that are robust to teacher error, that can accurately infer task details, and that are perceived to be intelligent. / Engineering and Applied Sciences

Industrial robots are delivering more and more manipulation services in manufacturing. However, when the task is complex, it is difficult to programme a robot to fulfil all the requirements because even a relatively simple task such as a peg-in-hole insertion contains many uncertainties, e.g. clearance, initial grasping position and insertion path. Humans, on the other hand, can deal with these variations using their vision and haptic feedback. Although humans can adapt to uncertainties easily, most of the time, the skilled based performances that relate to their tacit knowledge cannot be easily articulated. Even though the automation solution may not fully imitate human motion since some of them are not necessary, it would be useful if the skill based performance from a human could be firstly interpreted and modelled, which will then allow it to be transferred to the robot. This thesis aims to reduce robot programming efforts significantly by developing a methodology to capture, model and transfer the manual manufacturing skills from a human demonstrator to the robot. Recently, Learning from Demonstration (LfD) is gaining interest as a framework to transfer skills from human teacher to robot using probability encoding approaches to model observations and state transition uncertainties. In close or actual contact manipulation tasks, it is difficult to reliabley record the state-action examples without interfering with the human senses and activities. Therefore, wearable sensors are investigated as a promising device to record the state-action examples without restricting the human experts during the skilled execution of their tasks. Firstly to track human motions accurately and reliably in a defined 3-dimensional workspace, a hybrid system of Vicon and IMUs is proposed to compensate for the known limitations of the individual system. The data fusion method was able to overcome occlusion and frame flipping problems in the two camera Vicon setup and the drifting problem associated with the IMUs. The results indicated that occlusion and frame flipping problems associated with Vicon can be mitigated by using the IMU measurements. Furthermore, the proposed method improves the Mean Square Error (MSE) tracking accuracy range from 0.8˚ to 6.4˚ compared with the IMU only method. Secondly, to record haptic feedback from a teacher without physically obstructing their interactions with the workpiece, wearable surface electromyography (sEMG) armbands were used as an indirect method to indicate contact feedback during manual manipulations. A muscle-force model using a Time Delayed Neural Network (TDNN) was built to map the sEMG signals to the known contact force. The results indicated that the model was capable of estimating the force from the sEMG armbands in the applications of interest, namely in peg-in-hole and beater winding tasks, with MSE of 2.75N and 0.18N respectively. Finally, given the force estimation and the motion trajectories, a Hidden Markov Model (HMM) based approach was utilised as a state recognition method to encode and generalise the spatial and temporal information of the skilled executions. This method would allow a more representative control policy to be derived. A modified Gaussian Mixture Regression (GMR) method was then applied to enable motions reproduction by using the learned state-action policy. To simplify the validation procedure, instead of using the robot, additional demonstrations from the teacher were used to verify the reproduction performance of the policy, by assuming human teacher and robot learner are physical identical systems. The results confirmed the generalisation capability of the HMM model across a number of demonstrations from different subjects; and the reproduced motions from GMR were acceptable in these additional tests. The proposed methodology provides a framework for producing a state-action model from skilled demonstrations that can be translated into robot kinematics and joint states for the robot to execute. The implication to industry is reduced efforts and time in programming the robots for applications where human skilled performances are required to cope robustly with various uncertainties during tasks execution.

abstract: Geoscience educators commonly teach geology by projecting a photograph in front of the class. Geologic photographs often contain animals, people, and inanimate objects that help convey the scale of features in the photograph. Although scale items seem innocuous to instructors and other experts, the presence of such items is distracting and has a profound effect on student learning behavior. To evaluate how students visually interact with distracting scale items in photographs and to determine if cueing or signaling is an effective means to direct students to pertinent information, students were eye tracked while looking at geologically-rich photographs. Eye-tracking data revealed that learners primarily looked at the center of an image, focused on faces of both humans and animals if they were present, and repeatedly returned to looking at the scale item (distractor) for the duration an image was displayed. The presence of a distractor caused learners to look at less of an image than when a distractor was not present. Learners who received signaling tended to look at the distractor less, look at the geology more, and surveyed more of the photograph than learners who did not receive signaling. The San Antonio area in the southern part of the Baja California Peninsula is host to hydrothermal gold deposits. A field study, including drill-core analysis and detailed geologic mapping, was conducted to determine the types of mineralization present, the types of structures present, and the relationship between the two. This investigation revealed that two phases of mineralization have occurred in the area; the first is hydrothermal deposition of gold associated with sulfide deposits and the second is oxidation of sulfides to hematite, goethite, and jarosite. Mineralization varies as a function of depth, whereas sulfides occurring at depth, while minerals indicative of oxidation are limited to shallow depths. A structural analysis revealed that the oldest structures in the study area include low-grade to medium-grade metamorphic foliation and ductile mylonitic shear zones overprinted by brittle-ductile mylonitic fabrics, which were later overprinted by brittle deformation. Both primary and secondary mineralization in the area is restricted to the later brittle features. Alteration-bearing structures have an average NNW strike consistent with northeast-southwest-directed extension, whereas unaltered structures have an average NNE strike consistent with more recent northwest-southeast-directed extension. / Dissertation/Thesis / Ph.D. Geological Sciences 2011

Intelligent failure occurs when an entrepreneurial initiative falls short of its anticipated performance. It provides valuable new knowledge to the organisation and is recognised as an important factor in long-term corporate entrepreneurial success. This thesis is located within the domain of corporate entrepreneurship and entrepreneurial failure, and explores the various processes of intelligent failure. The specific aim of this thesis is to learn how organisations manage intelligent failure. Research takes an inductive approach with the predominant use of a qualitative methodology and, as part of a multiple case study strategy, research is carried out in six organisations operating in differing sectors within the UK. Findings indicate that the organisations often fail to manage intelligent failure. There is little evidence of a strategic approach to learning from failure and, where learning occurs, it is predominantly unstructured. This is significant because literature consistently argues that a structured process is required to manage learning from failure successfully. This research recognises that structured processes may be more effective than unstructured processes when looked at in isolation. However, this thesis argues that unstructured mechanisms do have inherent value. Therefore, when organisations develop failure management processes, a dual path may be considered, which might extract value from both systems as is contextually appropriate. This may enable organisations to maximise their ability to learn from failure. This thesis adds to existing management theory in the corporate entrepreneurship domain. In specifically focusing on the structured and unstructured forms within the process of intelligent failure, this thesis addresses a gap in current literature. It also adds to existing literature that centres on the practical management of the learning from failure process.

Needs that are shared by members of an organization can trigger an organizational learning process. To a large extent, needs are implicitly anchored in organizations and people can hardly articulate them. In this article, we present Bewextra, a method that allows for identifying hidden needs in organizations. Thereby, we trigger a knowledge conversion process, which is similar to Nonaka's SECI-spiral. In two case studies, we present how our Bewextra-process is applied to projects in educational contexts in Austria. In a first case study, we show that a combination of learning from past and future experiences extend the scope of the overall outcome. Since learning from future experiences requires a distinct environment (enabling spaces), we present a second case study. Here, we conducted a Bewextra-process with a large number of participants (n > 170), focusing on learning from future experiences.

Innovation processes require organizations to transcend current boundaries. These include not only technological as well as social limitations but "above all" the way we address the future. We are used to face the future with our existing knowledge and experiences from the past. This strategy, however, can hardly lead to knowledge off the beaten path. We therefore suggest a new learning approach for organizations, which enables to literally envision a desired future scenario and thereby, allows for the creation of radical new knowledge. We argue that the created knowledge yields a higher degree of novelty and radicalness. Along with an enhanced theory of learning including learning from the future, we present our empirical findings from comparing the outputs of Learning from an Envisioned Future and learning from the past. For this purpose, we use data from two organizational learning projects; one, which was conducted with a high school in Austria and another one, which was conducted with members of the Austrian Economic Chamber. Our findings from both case studies suggest that Learning from an Envisioned Future does produce significantly more paradigm challenging knowledge compared to the output gained from conventional learning from past experiences. We conclude that the combination of both learning sources may lead to best learning outcomes in organizations.

Learning from Demonstration (LfD) has been used in robotics research for the last decades to solve issues pertaining to conventional programming of robots. This framework enables a robot to learn a task simply from a human demonstration. However, it is unfeasible to teach a robot all possible scenarios, which may lead to e.g. the robot getting stuck. In order to solve this, a search is necessary. However, no current work is able to provide a search approach that is both simple and general. This thesis develops and evaluates a new framework based on LfD that combines both of these aspects. A single demonstration of a human search is made and a model of it is learned. From this model a search trajectory is sampled and optimized. Based on that trajectory, a prediction of the encountered environmental forces is made. An impedance controller with feed-forward of the predicted forces is then used to evaluate the algorithm on a Peg-in-Hole task. The final results show that the framework is able to successfully learn and reproduce a search from just one single human demonstration. Ultimately some suggestions are made for further benchmarks and development.

Purpose: Previous studies showed that combining learning based on experiences in the past with learning from an envisioned future scenario results in more innovative and radical ideas, as well as in a higher number of covered content domains. However, currently there is no holistic learning theory that integrates both sources of learning. The main purpose of this paper is to investigate whether it is possible to extend Bateson's theory of learning, to link these two learning sources in one coherent framework. Design/methodology/Approach: To answer this research question, the author draws on learning from an envisioned future, and tries to link it with the most important levels of learning in Bateson's framework. Findings: This paper contributes to the literature by attempting to link the important but still underexplored aspect of "learning from the future" to the complex and multifaceted work of Bateson. Given the fact that both sources of learning and experience yield a great potential to create new knowledge, this study outlines a possibility to include both sources into one learning theory. Research limitations/implications: This work provides the basis for further research in building a general holistic theory of learning to learn. Practical implications: On the individual level, the proposed approach can be easily applied with systemic coaching processes in general and coaching processes in the fields of developing an individual vision in particular. In the field of organizational learning, the awareness of different learning sources and different learning modes on the one hand and knowledge about the implementation of enabling spaces (PE-ba, FE-ba) to support these various learning modes on the other hand help organizations to generate new knowledge and create innovative and sustainable solutions, products and services. Originality/value: To the best of the author's knowledge, it is the first theoretical work that describes the integration of learning from past experiences and learning from future experiences in a methodological way.

This research focuses on the learning of a class of behaviors defined as high-level behaviors. High-level behaviors are defined here as behaviors that can be executed using a sequence of identifiable behaviors. Represented by low-level contexts, these behaviors are known a priori to learning and can be modeled separately by a knowledge engineer. The learning task, which is achieved by observing an expert within simulation, then becomes the identification and representation of the low-level context sequence executed by the expert. To learn this sequence, this research proposes FAMTILE - the Fuzzy ARTMAP / Template-Based Interpretation Learning Engine. This algorithm attempts to achieve this learning task by constructing rules that govern the low-level context transitions made by the expert. By combining these rules with models for these low-level context behaviors, it is hypothesized that an intelligent model for the expert can be created that can adequately model his behavior. To evaluate FAMTILE, four testing scenarios were developed that attempt to achieve three distinct evaluation goals: assessing the learning capabilities of Fuzzy ARTMAP, evaluating the ability of FAMTILE to correctly predict expert actions and context choices given an observation, and creating a model of the expert's behavior that can perform the high-level task at a comparable level of proficiency.

The gaming industry has reached a point where improving graphics has only a small effect on how much a player will enjoy a game. One focus has turned to adding more humanlike characteristics into computer game agents. Machine learning techniques are being used scarcely in games, although they do offer powerful means for creating humanlike behaviors in agents. The first person shooter (FPS), Quake 2, is an open source game that offers a multi-agent environment to create game agents (bots) in. This work attempts to combine neural networks with a modeling paradigm known as context based reasoning (CxBR) to create a contextual game observation (CONGO) system that produces Quake 2 agents that behave as a human player trains them to act. A default level of intelligence is instilled into the bots through contextual scripts to prevent the bot from being trained to be completely useless. The results show that the humanness and entertainment value as compared to a traditional scripted bot have improved, although, CONGO bots usually ranked only slightly above a novice skill level. Overall, CONGO is a technique that offers the gaming community a mode of game play that has promising entertainment value.