Title: The resiliency of robot swarms in collective decision making
Project description: Swarm Robotics involves the coordination of a large number of simple robots that communicate locally with their peers and the environment to perform a common task. Using behaviours similar to those observed in the natural systems, the robots need to make decisions, such as how to aggregate the objects, where to start the construction, or which direction to move towards in an environment. For most operations within a swarm, one of the essential abilities is to undertake collective decision-making, i.e., to agree on the best option from one of the possible alternatives in the environment. However, little attention has been devoted to studying the resiliency of the collective decision-making process from disturbances arising from malicious agents. In the project, we will computationally analyze swarm resiliency in collective decision-making while focusing on how various kinds of attack models affect its performance in terms of accuracy and speed. The success of this project can have an impact on helping the deployment of swarm robots safely in real-life applications.
Title: Federated learning in swarm robotics using a blockchain-based smart contract
Project description: Federated learning is used to distribute the training of a machine learning model across multiple devices (e.g., smartphones). The advantages of this technique include increased privacy, minimization of data exchange, and parallelization of computing resources. Specifically, federated learning means that the devices train a model locally and then exchange only the learned parameters (rather than the entire raw data). To aggregate the locally trained parameters, existing work mostly uses centralized servers.
This M.Sc. project aims at transferring federated learning to robot swarms—robot swarms are multi-robot systems consisting of a large number of autonomous robots whose swarm behaviour results from the self-organized interactions among the robots. In the proposed project, each robot in the swarm trains a model locally. However, in contrast to existing work that uses a central server to aggregate the parameters of the local models, we would like to use a decentralized and secure data structure, namely a blockchain that is maintained by the robot swarm.
Blockchain technology was originally developed for the decentralized currency Bitcoin inorder to send digital coins. Since 2015, there is an extension to the classic Bitcoin blockchain: Ethereum is a blockchain framework that allows programs to be executed in a decentralized blockchain network and to reach a consensus on the outcome of the programs. These programs that are stored on the blockchain, are called blockchain-based smart contracts (or just smart contracts, in short form). The goal of this project is to implement federated learning in a robot swarm using a smart contract. As a further challenge, this smart contract should be resilient to malfunctioning or even malicious robots: that is, robots that send bad parameters.
We developed a framework that combines the robot swarm simulator ARGoS and the blockchain framework Ethereum. The student should use this framework to explore, implement, and analyze a suitable scenario for federated learning in robot swarms
Title: An information market for social navigation in robots
Project description: In this project, social interactions between robots are the results of free-market dynamics where robots buy and sell information. The robots are tasked with moving back and forth between location A to location B, and receive a monetary reward for each successful round trip. However, the robots do not have a GPS and can remember the locations A and B only through the history of their past movements (that is, through odometry). Odometric estimates are subject to noise and therefore the robots can frequently get lost. In order to move between A and B, the robots need “information” about where A and B are. They can either obtain this “information” alone, which requires exploring the environment (in this case, the paid cost is time). Or, alternatively, they can obtain this “information” from other robots that recently visited location A or B, here the information is bought from other robots (in this case, robots incur a monetary cost).
The student is expected to extend an existing multi-agent simulator (in Python) which already have the basic mechanisms for movements and economic transactions. The extensions will include in the simulation a richer behaviour that will allow linking this study to real-world applications. Through the multi-agent simulations, different payment schemes will be tested, and the obtained results will beanalysed to understand under which conditions buying information is more advantageous than self-acquiring information (cost exploration > cost purchase), as well as which how the collective behaviour can be secured from selfish (non-collaborative) behaviour of a subset of robots.
Title: TraffiCoin -Improve traffic through blockchain technology
Project description: Self-driving cars will soon arrive in our cities and be the safest means of transport. A visionary approach consists in allowing cars to regulate traffic through economical transactions to gain or concede priority in a road intersection. The passenger will decide the level of urgency s/he has to arrive at the desired destination. The cars at an intersection will communicate with each other and, through an auction process with bids which depends on the passenger’s urgency, the car that wins the auction gets the priority to pass and in return, it will pay asmall price to the cars that wait to give way. These traffic dynamics can be regulated through the blockchain. Car queues at an intersection build up bidding power by submitting their bids into larger sums and getting priority at the intersection, in thisway the traffic gets more democratic and priority is given to larger groups. Additionally, more populated vehicles will acquire more bidding power and will move quicker in the city, giving concrete incentives in reducing cars with single passengers and hence contributing to a greener planet. Such a visionary approach will tackle a problem relevant to the emerging technologies of self-driving cars and blockchain technology in order to achieve traffic reduction. The student will be asked to design and run simulations based on existing traffic simulators which will be extended to interoperate with a blockchain. The central project goal consists in coding blockchain-based smart contracts that will improve traffic efficiency, and quantify their impact.
Title: Consensus decision making in human swarms
Project description: Collective decision making is one of the most important research topics in swarm robotics. Research conducted at IRIDIA has shown that, when faced with a choice between several options, a swarm of robots equipped with simple rules can collectively find the best option. In order for the swarm to succeed, there must be both a consensus and the selection of the best alternative. Individuals can see both the state of the world—to make estimates of options' qualities and choose the best alternative—and see other individuals' opinion.
An interesting observation has been that the decision making process is strongly altered when a subset of individuals are so-called uncommitted individuals who do not have access to information about the state of the world but can only follow others. In this case, we see that the opinion of a small but strongly opinionated minority will more often be chosen as the consensus opinion of the swarm.
In this master thesis project, you will analyse if this effect is also present when the robots are not controlled by simple algorithms, but by human participants. You will work together with researchers from the psychology faculty to recruit human participants, and run online multi-player experiments using our ‘HuGoS’ platform. You will implement the uncommitted individuals in the platform as artificial agents by coding in C# and using the Unity game engine. The results of this project might inspire new control algorithms for robot swarms.