Planning algorithms have applications in many fields such as robotics, logistics, and more.Planning for terrain following flights can be to utilize features of the terrain to minimizethe risk of detection. The similar problem of terrain avoidance is applicable to planningthe movement for survey or search vehicles, where low and fixed altitude may be important.Common problems that arise when planning for terrain following flights is that the dynamics ofthe vehicle are difficult to model, the state space is only represented in an approximate mannerand detailed calculations of the subject are computationally expensive. This work presents aplanning algorithm for the vertical component of terrain following flight paths using methods ofenergy, where the path itself is modelled as an elastic band deformed by virtual forces to followthe terrain. Using linear internal equations of motion for the elastic band, the computationalcomplexity can be kept low. Similar ideas have been used for planning algorithms in otherworks, but novel to the method presented here is that complicated limitations to the dynamicsof the vehicle can be treated in an effective manner. This is achieved by an adaptive linearcombination of different models for the internal elastic forces.

This thesis makes a number of new contributions to control and sensing for unmanned vehicles. I begin by developing a non-linear simulation of a small unmanned helicopter and then proceed to develop new algorithms for control and sensing using the simulation. The work is field-tested in successful flight trials of biologically inspired vision and neural network control for an unstable rotorcraft. The techniques are more robust and more easily implemented on a small flying vehicle than previously attempted methods.¶ Experiments from biology suggest that the sensing of image motion or optic flow in insects provides a means of determining the range to obstacles and terrain. This biologically inspired approach is applied to control of height in a helicopter, leading to the World’s first optic flow based terrain following controller for an unmanned helicopter in forward flight. Another novel optic flow based controller is developed for the control of velocity in hover. Using the measurements of height from other sensors, optic flow is used to provide a measure of the helicopters lateral and longitudinal velocities relative to the ground plane. Feedback of these velocity measurements enables automated hover with a drift of only a few cm per second, which is sufficient to allow a helicopter to land autonomously in gusty conditions with no absolute measurement of position.¶ New techniques for sensor fusion using Extended Kalman Filtering are developed to estimate attitude and velocity from noisy inertial sensors and optic flow measurements. However, such control and sensor fusion techniques can be computationally intensive, rendering them difficult or impossible to implement on a small unmanned vehicle due to limitations on computing resources. Since neural networks can perform these functions with minimal computing hardware, a new technique of control using neural networks is presented. First a hybrid plant model consisting of exactly known dynamics is combined with a black-box representation of the unknown dynamics. Simulated trajectories are then calculated for the plant using an optimal controller. Finally, a neural network is trained to mimic the optimal controller. Flight test results of control of the heave dynamics of a helicopter confirm the neural network controller’s ability to operate in high disturbance conditions and suggest that the neural network outperforms a PD controller. Sensor fusion and control of the lateral and longitudinal dynamics of the helicopter are also shown to be easily achieved using computationally modest neural networks.

Terrestrial robots must be capable of negotiating rough terrain if they are to become autonomous outside of the lab. Although the control mechanism offered by wheels is attractive in its simplicity, any wheeled system is confined to relatively flat terrain. Wheels will also only ever be useful for rolling, while limbs observed in nature are highly multimodal. The robust locomotive utility of legs is evidenced by the many animals that walk, run, jump, swim, and climb in a world full of challenging terrain. On the other hand, legs with multiple degrees of freedom (DoF) require much more complex control and precise sensing than wheels. Legged robotic systems are easily hampered by sensor noise and bulky control loops that prohibit the high-speed adaptation to external perturbations necessary for dynamic stability in real time. Low sensor bandwidth can limit the system’s reaction time to external perturbations. It is also often necessary to filter sensor data, which introduces significant delays in the control loop. In addition, state estimation is often relied upon in order to compute active stabilizing responses. State estimation requires accurate sensor data, often involving filtering, and can involve additional nontrivial computation such as the pseudo-inversion of fullbody Jacobians. This perception portion of the control burden is all incurred before a response can be planned and executed. These delays can prevent a system from executing a corrective response before instability leads to failure. The present work presents an approach to legged system design and control that reduces both the perception and planning aspects of the online control burden. A commonly accepted design goal in robotics is to accomplish a task with the fewest possible DoF in order to tighten the control loop and avoid the curse of dimensionality. However, animals control many DoF in a manner that adapts to external perturbations faster than can be explained by efferent neural control. The passive mechanics of segmented animal limbs are capable of rejecting unexpected disturbances without the supervision of an active controller. By simulating biomimetic limbs, we can learn more about this preflexive response, how the properties of segmented biological limbs foster self-stable passive mechanics, and how the control burden can be mitigated in robotic legged systems. The contribution of this body of work is to reduce the control burden of legged locomotion for robots by drawing on self-stabilizing mechanical design and control principles observed in animal locomotion. To that end, minimal templates such as Sensory-Coupled Action Switching Modules (SCASM), Central Pattern Generators (CPGs), and the Spring-Loaded Inverted Pendulum (SLIP) model are used to learn more about the essential components of legged locomotion. The motivation behind this work lies largely in the study of how internal, predictive models and the intrinsic mechanical properties of biological limbs help animals self-stabilize in real time. Robotic systems have already begun to demonstrate the benefits of these biological design primitives in an engineering context, such as reduced cost of transportation and an immediate mechanical response that does not need to wait for sensor feedback or planning. The original research presented here explores the extent to which these principles can be utilized in order to encourage stable legged locomotion over uneven terrain with as little sensory information as possible. A method for generating feedforward, terrain-adaptive control primitives based on a compliant limb architecture is developed. Offline analysis of system dynamics is used to develop clock-driven patterns of leg stiffness and attack angle control during late swing with which passive stance phase dynamics will produce the desired apex height and stride period to within 0.1 mm and 50 μs, respectively. A feedforward method of energy modulation is incorporated that regulates velocity to within 10−5 m/s. Preservation of a constant stride period eliminates the need for detection of the apex event. Precise predictive controls based on thorough offline dynamic modeling reduce the system’s reliance on state and environmental data, even in rough terrain. These offline models of system dynamics are used to generate a controller that predicts the dynamics of running over uneven terrain using an internal clock signal. Real-time state estimation is a non-trivial bottleneck in the control of mobile systems, legged and wheeled alike. The present work significantly reduces this burden by generating predictive models that eliminate the need for state estimation within the control loop, even in the presence of damping. The resulting system achieves not only self-stable legged running, but direct control of height, speed, and stride period without inertial sensing or force feedback. Through this work, the controller dependency on accurate and rapid sensing of the body height and velocity, apex event, and ground variation was eliminated. This was done by harnessing physics-based models of leg dynamics, used to generate predictive controls that exploit the passive mechanics of the compliant limb to their full potential. While no real world system is entirely deterministic, such a predictive model may serve as the base layer for a lightweight control architecture capable of stable robotic limb control, as in animal locomotion.

There are occasions when it is preferable that an aircraft flies asclose to the ground as possible. It is difficult for a pilot to predict the topography when he cannot see beyond the next hill, and this makes it hard for him to find the optimal flight trajectory. With the help of a terrain database in the aircraft, the forthcoming topography can be found in advance and a flight trajectory can be calculated in real-time. The main goal is to find an optimal control sequence to be used by the autopilot. The optimization algorithm, which is created for finding the optimal control sequence, has to be run often and therefore, it has to be fast. This thesis presents a terrain following algorithm based on Model Predictive Control which is a promising and robust way of solving the optimization problem. By using trajectory optimization, a trajectory which follows the terrain very good is found for the non-linear model of the aircraft.

<p>There are occasions when it is preferable that an aircraft flies asclose to the ground as possible. It is difficult for a pilot to predict the topography when he cannot see beyond the next hill, and this makes it hard for him to find the optimal flight trajectory. With the help of a terrain database in the aircraft, the forthcoming topography can be found in advance and a flight trajectory can be calculated in real-time. The main goal is to find an optimal control sequence to be used by the autopilot. The optimization algorithm, which is created for finding the optimal control sequence, has to be run often and therefore, it has to be fast. </p><p>This thesis presents a terrain following algorithm based on Model Predictive Control which is a promising and robust way of solving the optimization problem. By using trajectory optimization, a trajectory which follows the terrain very good is found for the non-linear model of the aircraft.</p>