Communication-Aware Robotics

Communication-Aware Robotics

Students: Arjun Muralidharan and Herbert Cai, Advisor: Yasamin Mostofi, UCSB

Multi-agent systems, Fundamentals of the co-optimization of sensing, communication, and navigation for successful task accomplishments under resource constraints Utilizing unmanned vehicles to enable new forms of connectivity: (top) robotic routers, (bottom) robotic beam forming Fundamentals of wireless channel predictability in realistic communication environments, optimal path planning for a connectivity-seeking robot

Research Summary

For more than a decade, our lab has been very excited about the possibilities created at the intersection of communications and robotics. Consider the general problem where a team of unmanned vehicles are given a task to do cooperatively. As such, maintaining some form of connectivity, among the nodes or to an outside node, can be crucial to the successful task accomplishment. Since path planning decisions of the robots can directly affect the connectivity, the unmanned vehicles need to take connectivity no.s into account when path planning. We then have the following important question: how should each robot make decisions that co-optimize navigation, communication, and sensing no.s properly, under resource constraints? In another example, consider using unmanned vehicles to enable a needed form of connectivity. More specifically, can a group of unmanned vehicles path plan to form a network that supports the needed connectivity? We have coined the term communication-aware robotics more than a decade ago to refer to such a body of problems created at the intersection of the two areas of robotics and communications. In order to properly solve such problems, tools from both areas of robotics and communications are needed. Furthermore, impact of realistic communication channels on robotic networks should be considered by considering factors such as multipath fading, shadowing, and path loss. In IEEE ICRA 2010 and IEEE TWC 2012, we have then shown how an unmanned vehicle can go beyond the typically-used but unrealistic disk models, and realistically predict the channel at unvisited locations, using a probabilistic framework (Gaussian process). Our approach has been extensively validated experimentally. This then opens up new possibilities for the co-optimization of communication and navigation in robotic operations, as we discuss next. See the publications for a complete list of our papers in this area.

Consider an unmanned vehicle that is not connected in its current spot. What are the statistics of the distance to connectivity (first passage distance)? In IEEE ACC 2017, we have mathematically characterized the PDF of the distance to connectivity, using stochastic differential equations and stochastic dynamic programming. Next, can this robot find a connected spot with minimal energy consumptions, without knowledge of the exact channel quality over the field, and in realistic channel environments that can experience multipath, shadowing, and path loss? In IEEE Globecom 2017 and submitted journal, we have shown how to optimally solve this problem (the first such result to the best of our knowledge) by formulating it as an infinite horizon Markov Decision Process problem. We have then proposed a path planner, using a game-theoretic framework, that asymptotically gets arbitrarily close to the optimal solution, but with a considerably less space complexity. It is noteworthy that our formulation is very general in this paper and addresses the general problem of minimizing the expected cost till success, where the success can be connectivity or other robotic goals.

Next, consider a number of unmanned vehicles. What are the fundamentals of utilizing mobility to enable new forms of network connectivity? In IEEE TCNS 2017, we have proposed a novel way of enabling connectivity, using a number of mobile robots that do cooperative beam forming. More specifically, we have shown how each robot should optimize its motion to move to a place better for cooperative robotic beam forming. This enables the unmanned vehicles to establish a strong link, with a minimal energy usage, although each individual one cannot establish the needed connectivity on its own. In IEEE TRO 2012, we have then shown how a number of unmanned vehicles can optimally path plan to act as robotic routers and establish a robust information flow between two otherwise remote nodes.

Next, consider a team of unmanned vehicles that is tasked with a mission in realistic communication environments that experience fading, shadowing, and path loss. The path of the robots affects both communication and sensing qualities. Thus, the communication, navigation, and sensing no.s should be co-optimized, especially in resource-constrained environments. Along this line, we have recently proposed (IEEE TCNS 2018) a new methodology for the co-optimization of sensing, communication and navigation objectives in a networked robotic operation, based on an optimal-control approach. More specifically, consider a team of robots tasked with collectively transmitting a given amount of data to a remote station, while operating in realistic communication environments that experience path loss, shadowing, and multipath fading. We have shown how to optimally design the load distribution, paths, and transmission power/rate schedules of the robots in a way that minimizes the total energy required for motion and communication. In IEEE TWC 2013, we have also considered this problem from an optimization theory perspective and shown underlying properties of the optimum motion-communication solution using KKT conditions. Overall, our methodology allows the robots to accomplish the given task in harsh environments, and with minimum resources in terms of total energy consumption, operation time, etc.

In IEEE TSP 2012 and IEEE TAC 2011, two specific application examples are considered: cooperative robotic surveillance (IEEE TSP 2012) and cooperative target tracking (IEEE TAC 2011), where some of our communication-aware design methodologies were tailored to these specific applications, to enable robust cooperative operation in realistic communication environments. Finally, the "To go or Not to go" problem was addressed in IEEE TCNS 2014. In many situations, an unmanned vehicle can incur motion energy to move to a better place for connectivity, or can increase its transmission power. While it is the general belief that motion will always be more expensive, we have shown (in IEEE TCNS 2014) under what conditions on the channel and motion parameters, it is more energy-efficient for the robot to incur motion energy to enable connectivity, and other what conditions it should simply increase its transmission power. As we have shown, there are several instances where it is more energy-efficient for the robot to incur motion energy and find a connected spot.

See the publications below for more details.

Back to top

Data and Code

Our research has resulted in several wireless channel measurements collected with our robots during our experiments. We have released some of our channel data, which can be useful for testing robotic approaches in realistic channel environments. You can find the released data here.

We have furthermore developed a realistic wireless channel simulator that can generate simulated 2D wireless channels by properly modeling the three underlying dynamics of a wireless channel, which are path loss, shadowing and multipath fading. You can find the corresponding codes here.

Back to top