Walking is a very efficient way of getting around and covering large distances. Due to impairments or in extreme conditions, such as carrying a heavy load, one might encounter difficulties while walking. In many cases, wheeled vehicles offer a solution. However, wheeled vehicles are often not suitable for indoor environments or heavy outdoor terrain. Furthermore, wheeled vehicles do not exploit the walking capabilities of the human. As an alternative, exoskeletons have been proposed. These exoskeletons fit around the human body as a portable mechanical suit. The effort and control needed to fulfill a task are shared by the human and the exoskeleton. Human physical effort is measured by metabolism. Metabolism can be measured by recording the intake and exchange of oxygen and carbon dioxide. Many different exoskeletons have been developed in recent decades. Recently experiments showed that walking metabolism can be reduced with an exoskeleton. The goal of this dissertation is to improve exoskeletons that reduce the metabolic cost of walking. One of the main difficulties in achieving this goal is the difficulty in determining in advance what the effect of the exoskeleton will be on the metabolic energy consumption of walking. As a consequence, the design process is characterized by trial and error. This dissertation contributes to improving the complete design process, which includes the modelling, the hardware and control design, and the evaluation of exoskeletons. Based on a literature review, three challenges were defined that facilitate a more systematic design approach for exoskeletons. These challenges are: • Improving knowledge of human–exoskeleton interaction • Improving exoskeleton hardware and control • Fast and detailed evaluation of exoskeleton concepts These challenges have been the cornerstones of the research described in this dissertation. IMPROVING KNOWLEDGE OF HUMAN-EXOSKELETON INTERACTION --Walking simulations The dynamics of human walking are highly non-linear. This has been shown in both simulation studies and experimental studies. The development of exoskeletons requires knowledge of this non-linear behavior. A way to predict this behavior is through biomechanical models. These models predict the kinematics, kinetics, muscle activation, and metabolism of walking (Geyer and Herr, 2010; van den Bogert et al., 2011). Until now, these models have not been used to predict walking with an exoskeleton. This dissertation makes a first attempt to use these models for exoskeleton design. The model developed by Geyer and Herr (2010) is used to simulate human walking with exoskeleton dynamics based on the exoskeleton by (Cain et al., 2007). The model of Geyer and Herr was used since it also has a model of the neuromuscular controller. This controller model has a relatively small number of parameters, which makes it suitable for optimization. Optimization of the control parameters showed that the walking model can adapt to exoskeletal walking. Some experimental trends were captured by the simulation study. However the model does not yet predict the quantitative results that can directly be used in the development process. --Empirical knowledge Since biomechanical models have insufficient accuracy to predict the metabolic cost of walking with an exoskeleton, an alternative solution must be found. One of these solutions is to use empirical data that has been obtained with studies with previous exoskeletons. This dissertation has further expanded this empirical knowledge. The XPED exoskeletons that are described in this dissertation are a realization of the exotendon concept of Van den Bogert (2003). This concept makes use of long elastic cables that run parallel to the human leg. These cables have a similar function to the long tendons that are observed in some animals that move very efficiently, like horses. The cables can temporarily store energy and redistribute energy over the joints. In simulation these exotendons reduce the human joint moments by 71 percent. This model-based prediction is based on the assumption that the joint angles do not change under the load and also the total joint torques stay are invariant. A second assumption is that a reduction in the human joint moments leads to a reduction in the walking metabolism. This dissertation contradicted both assumptions. Experiments with the Achilles exoskeleton, an active ankle exoskeleton, have shown that the joint angles are strongly influenced by the support provided by the Achilles exoskeleton. This should be taken into account when designing a support strategy for the exoskeleton. In the XPED and Achilles exoskeleton, the joint angle patterns were assumed to be influenced by the exoskeleton support. When the joint angles changed in the experimental studies, the support decreased. From this result it was concluded that the support should be robust against changes in the walking pattern. It is noted that in other exoskeletons (Malcolm et al., 2013; Sawicki and Ferris, 2008), the support was high despite the changes in the walking pattern. Still it is difficult to make an exact copy of the controllers of these exoskeletons for implementation in the Achilles exoskeleton since an exact description of the dynamics of these exoskeletons is not available. For the exoskeletons described in this dissertation, an exact description of the dynamics is included. The intention of this description is to make the results obtained with these exoskeletons more generally applicable. IMPROVING EXOSKELETON HARDWARE AND CONTROL Many exoskeletons are not powerful enough or are too heavy to be successful. This follows from regression equations comparing the results of different exoskeletons (Mooney et al., 2014a). In this dissertation, two design methods are presented that can be used to design exoskeletons that can generate much mechanical power and a relatively low weight. --Use of passive mechanisms If the mechanical power in exoskeletons is delivered directly by motors, these motors are relatively heavy. Analogous to mechanisms found in musculoskeletal systems, passive elements could be used to reduce the required motor power. For specific supports, it is even possible to design exoskeletons without motors. The previously mentioned XPED exoskeletons are an example of these passive exoskeletons. Passive elements can also be used in combination with active elements. An example in the human body is the combination of the soleus muscle and the Achilles tendon (Ishikawa et al., 2005). In this dissertation, a similar principle is applied in the Achilles exoskeleton. The Achilles exoskeleton supports the ankle push off. In this exoskeleton, a spring in series with an actuator is used. Temporarily storing energy in the spring can generate a higher mechanical peak power than the maximal motor power and reduce the energy consumption. --Numerical optimization The performance of the exoskeleton is determined by the interaction between many different components. It is difficult to see how changes in one component influence the functioning of other components. This dissertation solves this problem through modelling and optimization. A model of the exoskeleton is made that contains the (electro-)mechanical properties of exoskeletons. The dimensioning and choice for components can be acquired through optimization of the model. This principle has been applied in the design of the XPED and Achilles exoskeletons. --Improvement of exoskeleton control Walking is a cyclic motion. This dissertation has shown how this property of walking can be used to improve the force control of exoskeletons. The gait phase can be estimated with an adaptive frequency oscillator (AFO). Input to the AFO is a cyclic signal. In the case of walking, the hip angle or ground reaction force are suitable candidates. Based on the phase estimation cyclic signals can be estimated. The estimated signal is build up from primitive function. In this case, these are Gaussian functions. The amplitude of these signals is determined by a non-linear filter. The estimated signals can be used to improve tracking or to attenuate undesired dynamical effects. FAST AND DETAILED EVALUATION OF EXOSKELETON CONCEPTS --Improvement in gait analysis The human effort during walking and the change of human metabolic cost due to support with an exoskeleton is measured with respiratory analysis. This measure gives no insight in how changes in metabolic energy emerge. To get this insight, additional measurements are needed. Some of these measurements are kinetic and kinematic measures obtained from gait analysis. This analysis can, for example, be used to see how much mechanical power the human and the exoskeleton absorb and generate. Data is commonly acquired by tracking optical markers placed on the human body and measuring interaction forces with dynamometers such as force plates. Gait analyses are sensitive to errors and in the case of exoskeletal walking, the protocol is hindered due to occlusion of markers by the exoskeleton. The kinematic and kinetic acquired data is redundant. Current data analysis protocols do not make optimal use of this redundancy. This dissertation describes a generic method to process gait data based on an extended Kalman filter. The filter assumes consistent dynamics, and makes it possible to improve the accuracy of estimated joint angles moments, and estimate system parameters (e.g. segment lengths). The latter makes it possible to eliminate the need for palpation of anatomical landmarks. Since the method can be used in real-time, it can be used to evaluate the effects of changes in control settings of the exoskeletons while walking. --Exoskeleton testbeds The development of new hardware to evaluate new exoskeleton concepts is very time consuming. It would therefore be beneficial to be able to test multiple concepts on one platform, an exoskeleton testbed. This requires some flexibility in the hardware and control. Also the dynamics of the exoskeleton should be well defined. This makes it possible to generalize the knowledge that is obtained with exoskeletons and use it in new exoskeleton designs. In this dissertation, two exoskeletons are described that could serve as a testbed. The Achilles exoskeleton is an autonomous exoskeleton for support of the ankle. The Achilles exoskeleton is force controlled and different controllers can be implemented on the exoskeleton. Secondly, this dissertation evaluated how existing rehabilitation robots can be used to simulate the design of new exoskeletons. This dissertation specifically focuses on attenuation of the existing exoskeletons dynamics and improvement of the tracking. CONCLUSION The goal of this dissertation was to improve exoskeletons that reduce the metabolic cost of walking. The research has not directly led to such new exoskeletons. One of the main causes is the difficulty of predicting with sufficient accuracy the effect of an exoskeleton on the walking kinetics, kinematics, and metabolism. Some biomechanical models that might be suitable for this are available and have also been used in this dissertation. However, these models have not been validated. Therefore this dissertation paid special attention to the evaluation of exoskeletons to make these validation studies possible. Altogether, this has led to new methods to model, design, and evaluate exoskeletons. Hopefully, these methods will be valuable tools for the design of future exoskeletons.