CHAPTER 8
Conclusion
Our work has broadly explored the applications of quantum mechanisms to robotics, consolidating the previously fragmented discussions across disciplines, each of which contributes a piece to the emerging field of quantum robotics. While much of our exploration of possibilities has been theoretical (as implementations of relevant quantum engineering are still far from practical), we have suggested some of the key ways that developments in quantum science and engineering will impact the world of robotics, much as they have already begun impacting other fields.
Our journey into quantum robotics began with fundamental quantum mechanics principles, exploring the qubit representation of an atom and the representation’s capabilities to develop potentially breakthrough hardware possibilities for robotics. Advances in the underlying hardware storage and power systems of robots could have a tremendous impact on functionality, as these have classically been some of the bottlenecks of embedded systems design. Qubit representation of data can hypothetically be more memory and energy efficient than classical computer binary bits. By storing data within quantum superposition, the quantum memory may scale to represent exponentially more data in the same number of bits than classically possible. By manipulating bits approaching the Landauer Limit of energy, quantum circuits can theoretically use millions of times less energy than classical computer circuits.
If the quantum hardware upgrades for robots sound useful, the quantum software upgrades inspire even more cause for excitement. Quantum parallelism, Grover’s algorithm, and adiabatic optimization are key approaches to providing asymptotic speedup for simple, naïve brute force search in a parameter space—a speedup in one of the most fundamental computer science problems. These methods offer quadratically better runtimes for a multitude of classical robotic algorithms that rely on search as a primary aspect of their design. When extended to robotic planning algorithms, Grover Tree Search for both the uninformed and informed cases similarly exhibits possible speedups over classical approaches.
In addition to enhancements in search and planning, quantum mechanisms may provide a boost to robot learning. Quantum robotic agents can potentially learn quadratically faster than classical agents using a hypothetical “quantum active learning” algorithm. This enhancement in robot learning could allow robots to comprehend and perform well in more complex and unstructured environments than previously possible. Quantum robots may not be able to query their environment in superposition, but they are able to compute over large data sets of environment percepts in fewer iterations than a classical robot and thereby reason more robustly over its uncertainty factors.
Our broader survey of quantum machine learning also illustrated the many speedups a significant proportion of machine learning algorithms are expected to experience when deployed in quantum media. Being able to learn from data faster will allow for more sophisticated perception algorithms in which a quantum robot could more quickly learn complex models of an environment. Given the complexity of many modeling problems in robotics (such as that of manipulation and contact force dynamics), the enhancements of quantum machine learning will be welcome. Robots will likely be able to perceive patterns in data faster and more robustly than previously possible, a boon for more precise sensing and control.
In addition, many machine learning algorithms become more general in the quantum world, not requiring assumptions such as convex loss functions in SVMs or approximations such as contrastive divergence in Deep Boltzmann Machines. The generality allows for more modeling power of algorithms and added capability to learn more complex real-world functions than classical robots. A deeper, more intricate representation of an environment would allow robots to function better in a world governed by immense complexity.
Models discussed in this book also shed light on an interesting new possibility: robots operating in quantum environments or controlling quantum phenomena. qMDPs and QOMDPs extend the classical capabilities of MDPs and POMDPs respectively, facilitating planning and control strategies for robots manipulating quantum environments. HQMMs extended HMMs to allow robust filtering of quantum percepts. In addition, our survey explored a variety of control models such as Bilinear Models and Markovian Master Equations that extend robot controls primitives to the quantum world.
The actual implementation of quantum robotic technology is very much in its infancy. After all, we have yet to fully develop a working quantum computer, let alone a quantum robot that uses one. However, several components of quantum robots have been prototyped. For example, different types of optical artificial neural networks have been considered, different adiabatic optimization approaches have been explored for various machine learning algorithms, and many proposals are on the table for implementing active learning algorithms, HQMMs, and other such frameworks. The commercial implementation for quantum computing which has made the most progress, D-Wave, was explored as a case study in our book, highlighting various frameworks for defining a quantum computer, and providing a benchmark for understanding how far the field has progressed and what implementation challenges remain for the field of quantum robotics to achieve its full potential.
This literature review is just the beginning for the community. Much work clearly remains to be done before one can construct a quantum robot. Also (as Aaronson [2015] notes), many of the algorithms and methods discussed in this survey have been shown to work only in mathematical theory, and, even then, are limited in their scope and application. They cannot yet be widely applied to the point that, as some newspapers erroneously suggest, quantum computers will be able to replace all classical ones overnight. Thus it is imperative that one understands the “fine print” carefully articulated in the research papers which comprise the field to understand the mathematical preconditions necessary to be met before a particular algorithm or method can be applied.
While still in its infancy, the field of quantum robotics still holds great promise. A quantum robot is likely to possess interesting properties, some of which we’ve detailed in this book. As we better understand the underlying components needed to build a quantum robot, our understanding of the properties of a quantum robot will increase as well. Current theory suggests that, compared with the robots of today, quantum robotics will be able to interact with more complex environments, operate with greater energy and memory efficiently, learn faster and more robustly, and have the ability to manipulate and control quantum phenomena. Such robots could potentially contribute to productivity in the workplace, assist in the development of scientific research, and improve our quality of life. The emerging world of quantum robotics promises to be an exciting one. We hope our work has inspired you to delve deeper into the academic literature and learn more about it.