Quantum Computing in Robotic Science and Applications

Gazing at quantum computing-powered applications of the future, we frequently look to the technology’s potential to solve computationally-intensive mathematical problems, which could lead to breakthroughs in drug discovery, logistics, cryptography, and finance.?

But what could be the impact of combining quantum computing with increasingly advanced, AI-driven robotics??We will structure our remarks according to the classical sense-think-act cycle extended by an over all observe action that represents system diagnosis.

Sense: Perception, Vision, and Sensor Data Processing

Today’s autonomous robots, such as those on use in the manufacturing production line, often rely on computer vision in order to make assessments about their environment. Deployed to tasks which can include identifying faults on part of a jet engine, for example, these systems have to compute results on millions of pixels.

As visualized in Fig.2 the field is categorized into quantum image representations(QIRs), transformations, applications, and algorithms, some of which are also required for robotic perception. For instance, registration algorithms transform images, recorded by different sensors, indifferent times, with different depth, or from different viewpoints, into one common coordinate system.

Quantum image processing (QIMP) could emerge as a way of better understanding visual information, and store and process this data more efficiently, with quantum properties such as entanglement and parallelism.

The basic idea is that properties of an image, like the colors at certain positions, can be encoded as qubit-lattices and visualized into quantum image representation, transformations, applications and algorithms required for robotic perception.

QIMP only deals with two-dimensional images presently, which is not sufficient for robotic perception which deals with input from multiple sensors in order to locate and identify objects and environments. The report suggests QIMP will continue to develop well beyond traditional equivalents.

Think: Traditional Artificial Intelligence in Robotics

Traditional AI, in contrast to modern machine learning approaches, is based on formal knowledge representations (rules and facts) and algorithms in order to optimize the robot behavior or mimic intelligent behavior.?

AI-applications are frequently used in robotics, like path planning, the deduction of goal-oriented action plans, system diagnosis, the coordination of multiple agents, or reasoning and deduction of new knowledge. Many of these applications use variations of uninformed (blind) or informed (heuristic) search algorithms, which are based on traversing trees or graphs, where each node represents a possible state in the search space, connected to further follow-up states.?

Quantum computing can work as an alternative for almost every search algorithm used in robotics and AI applications and reduce complexity. combinatorial search algorithms can be formulated as quantum algorithmic problem by applying Grover’s algorithm, or quantum annealing, which reduces the complexity tremendously. For graph search algorithms there is also a quantum alternative based on quantum random walks.

Another group of AI algorithms deals with decision making under uncertainty by means of stochastic processes. This is usually accomplished by using Bayesian networks or Markov chains which are basically graphs where transitions between states are described by a stochastic properties.

Quantum algorithms for these applications are already formulated, like quantum Markov chain and quantum Markov processes, which replace the classical definitions of probability with quantum probability. Simulated annealing, genetic algorithms and some Monte Carlo methods can be interpreted as stochastic processes with a finite state space and can thus profit from a quantum formulation, which provides a quadratic speedup inmost of the cases. Simulated annealing as well as genetic algorithms are used in several approaches to tackle robot trajectory planning, which may as well profit from the quantum representation.

Act: Kinematics and Dynamics

Efforts have been made in recent years to solve classical robotic tasks using AI as an alternative. In the quantum realm, quantum neural networks could help solve problems relating to kinematics, or the mechanical motion of robots.?

The report states how the two levels of control in robotics, abstract task-planning, and specific movement-planning— which are currently solved separately— can be solved in a more integrative way with quantum computing.?

Fig.3 shows a qualitative classification of basic optimization options that can be applied to the design of a manipulator and its kinematics control.

The manipulator placement problem and the optimization of the manipulator design are even more complex in terms of the search space size. In addition, classical approaches fail completely in finding the optimal solution to the overall problem, which takes all optimization options into account at the same time. Same applies if the exactness of mathematical modeling is extended to the dynamic case, where, e.g., the moments of inertia of manipulator parts and joint friction are taken into consideration. There are opinions in the robotics community that this problem class can only be handled with quantum reinforcement learning, as with models that learn to improve themselves. Another potent method to get such combinatorial problems under control seems to be hybrid quantum-classical algorithms. Furthermore, the use of quantum computers in a multi-state operation appears to be useful to find the global solution of complex optimization problems. For this purpose, a first quantum computer resource for the generation of start solutions and a second quantum computer resource for the finding of optimal solutions are used.

Observe: Diagnosis and Data Mining

Even carefully-designed and manufactured robots encounter faults as degradation takes hold over time, or owed to an incomplete knowledge of an operating environment. Methods to detect and specify faults are essential, yet diagnoses of problems— finding the cause of the discrepancy between observed and expected behavior— can be a complex task.

The process of extracting information and identifying patterns from large data sets, data mining is a valuable tool for knowledge discovery in order to diagnose systems, such as identifying sources and reasons for erroneous robot behavior, by mining a robot’s log files.

Quantum algorithmic approaches can be well-suited to tackling the heavy data mining and analysis involved, making this crucial process— and the advancement of robotics technology— more efficient as a result.