Courbe d'apprentissage | EPFL Graph Search

thumb|Exemples de courbes d'apprentissage dans le secteur de l'énergie. Une courbe d'apprentissage est une représentation graphique de l'apprentissage avec l'expérience. Cette courbe décrit notamment la relation entre la production cumulée d'une entreprise et les quantités de facteurs nécessaires pour produire chaque unité. Une règle informelle dit que « lorsqu'on double son expérience industrielle, on arrive à réduire ses coûts de 30 % ». Enjeux de la courbe d'apprentissage Empirique comme la plupart des lois scalantes, elle mérite tout de même qu'on s'y arrête : si elle a pu être formulée, en effet, c'est qu'elle ne heurtait pas outre mesure l'expérience de ceux qui l'invoquaient, qu'il s'agisse de fabriquer des briques, des semiconducteurs, des autocommutateurs, des pull-overs ou du courant électrique. Lorsqu'on sait qu'on fera dix fois plus de quelque chose, on peut consacrer un peu de temps à étudier la façon dont on pourra le faire mieux, plus vite et moins cher — c

Supernumerary Robotic Arm for Three-Handed Surgical Application: Behavioral Study and Design of Human-Machine Interface

Elahe Abdi

In surgical to industrial manipulation, the operator needs assistance for tasks requiring more than two hands. Teamwork may be the source of errors and inefficiency, especially if the assistant is a novice or unfamiliar with the main operator. The need for assistance may become problematic in case of lack of human resources e.g. in emergency surgical cases in the late hours of the night. Our objective is to improve the surgeon's autonomy and dexterity by a robotic arm under his own control. Although a number of robotic instrument holders have been developed, the best way to control such devices is still an open question. No behavioral study has been conducted on the best control strategy and human performance in three-handed tasks. We have selected the foot for commanding the third arm on the basis of a literature review. A series of experiments in virtual environments has been conducted to study the feasibility of this choice. The first experiment compares performance in the same task using two or three hands. Results show that three-handed manipulation is preferred to two-handed manipulation in demanding tasks. The second experiment investigated the type of tasks to be aimed in three-handed manipulation and the learning curve of users. Moving the hands and a foot simultaneously in opposite directions was perceived as difficult compared to a more active task with liberty in choosing the limbs coordination. Limbs were moved in parallel rather than serially. The performance improved within a few minutes of practice. Also, the sense of ownership improved constantly during the experiment. Two other experiments were aimed at handling the endoscope in laparoscopic surgery. Surgeons and medical students participated in these studies. Residents had a more positive approach towards foot usage and performed better compared to more experienced surgeons. This proves that the best training period for surgeons to use a foot controlled robotic arm is during their residency. A realistic virtual abdominal cavity has been developed for the last experiment. This had a positive influence on the participants' performance and emphasizes the importance of using a familiar context for training such a "three-handed surgery". Finally, two different foot interfaces were developed to investigate the most intuitive third arm commanding strategy. A robotic arm is hence controlled by the foot's translation or rotation in one interface (isotonic interface), and by force or torque in the other one (isometric interface). An experimental behavioral study was conducted to compare the two devices. Isometric rate control was preferred to isotonic position control due to the lower physical burden and higher movement accuracy of the robot. It was shown that the proposed device for isometric rate control could be used for intuitive control of four DoFs of a slave robotic arm. This thesis is the first step in a systematic investigation of a three-handed manipulation, two biological hands and a foot controlled robotic assistant. Findings suggest a high potential in using the foot to become more autonomous in surgery as well as other fields. Users can learn the control paradigm in a short period of time with little mental and physical burden. We expect the developed foot interfaces to be the basis of future development of more intuitive control interfaces. We believe that foot controlled robotic arms will be commonly used in various surgical as well as industrial applications.

EPFL2017

Development of an Energy-Economic Model with Endogenous Technological Progress and Feasibility Study of CCS Systems

Takafumi Usui

The main objective of this research is to develop a bottom-up energy-economic model considering endogenous technological development. The designed model analyzes the feasibility of CCS technologies in the Japanese electricity market to derive an optimum carbon reduction scenario. Two factors, a learning curve based on learning-by-doing and public R&D investment, precede technological progress. The analysis is calculated with a set of scenarios, which is based on alternative assumptions for technological characteristics: chemical and physical carbon absorptive technology. From modeling estimation, we conclude that technological progress reduces the generation costs of conversion technologies with CCS, as a CCS system acquires an additional unit of installation. Generation cost with chemical absorption remarkably reduces its marginal unit cost through a learning mechanism. The supply fraction from a gas-fired power plant increases over the analytical time period. The introduction of CCS reduces carbon emission level 17% compared to the baseline scenario in 2050. Technological progress has little impact on the total system costs; however, learning-by-doing pushes the introduction of CCS into the market rather than into a R&D activity. Research and development efforts in the private sector or knowledge spillover are not modeled in the study; however, they have the potential to contribute to the mitigation of carbon emission as well.

Wiley-Blackwell2013

The Impact of Learning-by-doing and Learning-by-searching on the Diffusion of Renewable Energy Technologies in the Japanese Electricity Sector

Takafumi Usui

The mitigation of climate change is thought by many to be a significant environmental concern. In order to mitigate the bulk of carbon emissions from the electricity sector, large market penetration of renewable energy technologies is a key research issue. The goal of the paper is to address the timing of public R&D investment for renewable technologies in the Japanese electricity sector. The model is cast as a dynamic bottom-up energy-system model and numerically optimized, minimizing a total system cost subject to an accumulated carbon emission constraint. Autonomous energy-efficiency improvement and two-factor learning curves describe the technological development process for each energy technology. With the two-factor learning curve, the model shows that the cost of electricity of renewable technologies can be reduced by up to 43% in the year 2100 if there is no constraint on public R&D budget. Renewable energy sources requires substantial R&D support in their initial phase of development; however, the investment pattern would switch to a path of reducing the amount after 2050. The budget of R&D investment is strictly limited; thus, an important policy objective is to design a dynamic investment schedule that will induce sustainable innovations for green technologies.

2015

Supernumerary Robotic Arm for Three-Handed Surgical Application: Behavioral Study and Design of Human-Machine Interface

Elahe Abdi

EPFL2017