Robots of a feather, flock together, rock together

Can robots swarm together – as smartly as bees and birds? How do variables, conflicts, and unintended effects play out in such swarms? What about byzantine robots?

GITAM Deemed to be University-Hyderabad has recently bagged the Indian National Science Academy (INSA) 2023 Visiting Scientist Fellowship. Dr. Kummari will now be conducting research on "AI-Inspired Optimisation for Swarm Robotics" at the Advanced Computing and Microelectronics Unit of the Indian Statistical Institute (ISI) in Kolkata. The chief aim is to investigate the effectiveness of AI-inspired optimisation approaches in swarm robotics applications, which involve collaborative work among large groups of robots to achieve shared objectives. What is exactly happening and how will it matter to our world beyond the ‘Maths’. We decode here.

Congrats Dr.Krishna. What are your goals next? How relevant is this subject for real world use?

I want contribute to the optimisation theory and its applications, as well as apply AI to real-world problem-solving. Swarm robotics has a lot of applications in medical, manufacturing and agriculture areas. Just like how birds work towards one goal in a flock, like foraging food or escaping bad weather—there are goals which can be optimised with such models.

What exactly are you researching on here: and how does it shape algorithms in swarm robotics?

In fact, optimising the behaviour of each robot in a swarm is hard because swarm robotics is decentralised and needs spontaneous behaviour from the group as a whole. Optimisation methods that are based on AI can be a crucial part of making swarm robotics apps work better. Optimisation techniques that are based on AI can be used to shape algorithms for swarm robots. These programmes focus on how to change the behaviour of individual robots so that the behaviour of the swarm as a whole is what is wanted.

Do you lean towards linear or non-linear method?

In the field of Swarm Robotics, the choice between linear and nonlinear optimisation methods depends on the job at hand and the kind of behaviour the swarm is supposed to have. Both linear and nonlinear methods have advantages and disadvantages, and the choice is often based on how complex the problem is, how much computational power is available, and what level of performance is needed.

In swarm robotics, problems often come from the fact that the relationships between individual robots, the surroundings, and the behaviour of the swarm as a whole are not always linear. Because of this, nonlinear optimisation methods are often the best way to describe and improve swarm behaviour. But it's important to look at the characteristics of the problem and the amount of computing power available to figure out which method (linear or nonlinear) will work best in a given case. We focus on linear and nonlinear methods in this work.

Can you share more about the importance of optimisation in Swarm Robotics?

Optimisation is a crucial part of Swarm Robotics, which is the study of building and controlling groups of simple robots to do complex tasks by interacting and working together. When different parts of swarm robots are optimised, it can lead to better performance, efficiency, and reliability in reaching goals. Optimising is important in swarm robots because of many reasons. Swarm robots often work together to reach a common goal, like exploring an environment, making patterns, or finishing a specific task. Optimisation helps find the best ways for the swarm to work together to reach these goals with the least amount of overlap, collisions, and lost effort. Also, Swarm robots often work in environments that are uncertain and complicated, where sensor noise, communication problems, and other unexpected things can happen. Optimisation can help come up with solid plans that take into account uncertainty and make sure the swarm can recover well from disruption. So they are significant from the aspects of task efficiency and robustness. Also, tuning different factors and settings is part of making swarm algorithms that work well. Optimisation methods can automate and simplify this process, speeding up algorithms' development and making them work better.

Can you elaborate on the ways to improve performance that AI has inspired?

PSO is an optimisation method based on how birds or fish act in a group. In the field of swarm robotics, PSO can be used to optimise the features of individual robots, like how they move or talk to each other, to get the desired behaviour at the swarm level. Each particle in the swarm is in a position that represents a possible answer, and the particles change their positions based on what they have learned and what their neighbours have learned. Then there is Ant Colony Optimisation (ACO).  This method is based on how ants look for food. It can be used in swarm robotics to figure out the best paths or ways for each robot in the group. By modelling how pheromones stick to and leave surfaces, ACO can help robots find good routes and avoid traffic.

Next is Genetic algorithms – they are like evolution and natural selection. They can be used to change the features or behaviours of each robot in a swarm as generations pass. This method can help find effective ways to get a multitude to act in a certain manner. Apart from these there are ways like Artificial Neural Networks (ANN, Reinforcement Learning (RL) and hybrid approaches. How well these AI-inspired methods to optimising algorithms for swarm robotics work depends on many things, such as the problem at hand, the characteristics of the swarm, and the emergent behaviours that are wanted. Experiments and simulations are often needed to fine-tune these methods and determine how well they work in different situations.

Can faster, cheaper computational tools help in optimisation in a big way? 

Yes, faster and cheaper tools for computing can have a big effect on optimisation in many areas. To find the best answer, many optimisation algorithms must look through a search space. With faster computing tools, different candidate solutions can be evaluated more quickly. This lets optimisation algorithms search the search area more thoroughly and find better solutions. Plus, there are areas like Iterative Algorithms, Hyperparameter Tuning, and simulation and modelling.

In short, faster and cheaper computing tools can greatly affect optimisation by letting more of the solution space be explored, making iterations more quickly, letting more significant problems be solved, and giving more accurate and efficient solutions in many different domains.

Can real world application of swarm robotics pose conflicts in optimisation—specially in terms of the choice of decision variables and constraints?

In terms of selecting decision variables and constraints, deploying swarm robotics in the real world can pose optimisation challenges. Utilizing emergent behaviours resulting from local interactions between individuals, swarm robotics involves the coordination of a large number of relatively simple robots to execute complex tasks. Due to the numerous trade-offs and conflicts that arise from the various objectives and constraints involved, optimising swarm behaviour can be difficult.

Please expand on this.

When optimising the behaviour of a cluster of robots, there may be a trade-off between ensuring the safety of the robots and their surroundings and completing tasks quickly. For instance, optimising for speed in a warehouse with multiple robots transporting products could increase the risk of collisions. To attain a balance between safety and efficiency, it would be necessary to identify parameters that maintain this balance.Optimising for resource efficiency, such as battery usage or energy consumption, can conflict with the goal of completing tasks quickly or thoroughly. For example, a swarm of drones entrusted with environmental monitoring may need to find a balance between conserving battery life and adequately covering a large area.

Also, Swarm robots may need to find a balance between exploring new areas to acquire data and exploiting known areas to complete tasks efficiently. The selection of decision variables and constraints may be influenced by whether the swarm should prioritize exploration to discover new opportunities or exploitation to maximize task completion. Similarly, there are facets like Robustness versus Complexity or Adaptability versus Predictability.

In each of these circumstances, selecting the appropriate decision variables and constraints requires careful consideration of the application's objectives, the specific trade-offs involved, and the level of flexibility required to manage a wide range of real-world situations. Frequently, competing objectives necessitate a multi-objective optimisation strategy, in which different objectives are weighed against one another to find a suitable compromise.

Recently, some researchers have tried to fix the ‘byzantine robot’ problem with preemptive solutions in collective intelligence. Can such approaches help in fighting harmful robots?

Like any technology, such robots can be used as per the intent of the human being using them. Sothey can be used by the Army or by a bad guy – it depends on the goal one has behind using such tools. Cryptography can be quite useful in combatting threats of ill-aimed usage of such robots.

Can we get any templates from Biomimicry in swarm robotics—like beehives or migratory birds?

Biomimicry is an exciting idea in which natural designs and processes are used to solve human problems. In the field of swarm robots, examples from nature, like beehives and migrating birds, can be used as models. Like—Beehives show how well people can adapt to changing situations. This ability to change can be imitated by swarm robots, which can change their behaviour and jobs based on the task and their surroundings. As to migratory birds, they talk to each other to share knowledge about where food is and how the weather is. Communication can help a group of robots find their way and avoid obstacles. Robots can share information about their surroundings and possible routes to reach their goal as quickly as possible. If one bird gets tired or falls behind, another can take its place in a group of migrating birds. In swarm robotics, redundancy can be achieved by having multiple robots that can do the same job. This ensures the swarm can keep doing its job even if some robots break down.

Swarm robotics can use beehives and migrating birds as models to improve communication, coordination, adaptability, and total efficiency. Keep in mind that these ideas are helpful, but putting them into practice will require thinking about technical difficulties and the goals of the swarm robotics project.

How crucial is pure research and such projects for contributing to the Make-in-India movement?

A lot. Science and Maths form the foundation of real-world engineering progress. Putting focus on Maths and Computer Science can build a lot of capabilities for the end goal of strengthening India’s prowess.

Dr. Krishna K, Assistant Professor, GITAM Deemed to be University-Hyderabad

By Pratima H

pratimah@cybermedia.co.in