Some of our current projects include:
Ad-hoc Teamwork
In real multi-agent systems, agents may need to coordinate with others without having pre-programmed coordination methodologies, nor fully knowledge of how these other agents may operate. Hence, they must learn on-line models of other agents in the system, for on-line decision making. Therefore, we are investigating novel learning and reasoning techniques for such challenging problems.
Paper Examples
- E. S. Yourdshahi, M. A. C. Alves, L. S. Marcolino, P. Angelov. “On-line Estimators for Ad-hoc Task Allocation: Extended Abstract”. In Proceedings of the 19th International Conference on Autonomous Agents and Multiagent Systems (AAMAS 2020), May 2020. Download paper
- E. S. Yourdshahi, T. Pinder, G. Dhawan, L. S. Marcolino and P. Angelov. “Towards Large Scale Ad-hoc Teamwork”. In Proceedings of the 3rd International Conference on Agents (ICA 2018), July 2018. Download paper
On-line Learning and Planning in Swarm Environments
A large number of agents or robots, also known as swarms, have a great range of applications, such as patrolling, mapping, foraging, rescue, etc. They usually follow simple rules, but the system exhibit complex behavior, and its decentralised nature makes them robust and fault tolerant. However, a swarm may not be under an agent’s control, and could even be an agent’s antagonist. In these scenarios, they usually follow unknown algorithms, making it hard to predict their behavior. Therefore, we are investigating novel large scale on-line learning and planning techniques in order to aid an agent to handle such challenging situations.
Paper Examples
- L. Pelcner, S. Li, M. A. C. Alves, L. S. Marcolino, A. Collins. “Real-time Learning and Planning in Environments with Swarms: A Hierarchical and a Parameter-based Simulation Approach”. In Proceedings of the 19th International Conference on Autonomous Agents and Multiagent Systems (AAMAS 2020), May 2020. Download paper
Semantic Hyperlapse Videos
The rapid increase in the amount of published visual data and the limited time of users bring the demand for processing untrimmed videos to produce shorter versions that convey the same information. Despite the remarkable progress that has been made by summarization methods, most of them can only select a few frames or skims, which creates visual gaps and breaks the video context. In this project, we are investigating novel methodologies based on reinforcement learning formulations to accelerate instructional videos. Our approaches can adaptively select frames that are not relevant to convey the information without creating gaps in the final videos. Our agents are textually and visually oriented to select which frames to remove to shrink the input video.
Paper Examples
- W. De Souza Ramos, M. M. Silva, E. R. Araujo, L. S. Marcolino, E. R. Nascimento. “Straight to the Point: Fast-forwarding Videos via Reinforcement Learning Using Textual Data”. In Proceedings of the 2020 IEEE Conference on Computer Vision and Pattern Recognition (CVPR 2020), June 2020. Download paper
Novel Rendering Engines for Data Hungry Computer Vision Models
Large-scale synthetic data is needed to support the deep learning big-bang that started in the recent decade and influenced almost all scientific fields. Most of the synthetic data generation solutions are task-specific or unscalable while the others are expensive, based on commercial games, or unreliable. In this project, new rendering engines are being investigated. Photo-realism, diversity, scalability, and full 3D virtual world generation at run-time are key aspects that we want to consider. We aim at providing clean, unbiased, and large-scale training and testing data for various computer vision tasks.
Paper Examples
- A. Kerim, L. S. Marcolino, R. Jiang. “Silver: Novel Rendering Engine for Data Hungry Computer Vision Models”. In the 2nd International Workshop on Data Quality Assessment for Machine Learning, August 2021. Download paper
Music Generation in Digital Games
The concept of gamified interactive models has been widely approached in order to engage users in many fields. In fields such as HCI and AI, however, these approaches were not yet employed for supporting users to create different forms of artworks, like a musical corpus. While allowing novel forms of interactivity with partially-autonomous systems, these techniques could also foster the emergence of artworks not limited to experts. Hence, in this project we investigate the concept of meta-interactivity for compositional interfaces, which extends an individual’s capabilities by the translation of an effort into a proficiency. We are developing novel systems that enables non-experts to compose coherent musical pieces through the use of imagetic elements in virtual environments.
Paper Examples
- M. Escarce Junior, G. R. Martins, L. S. Marcolino, Y. T. dos Passos. “Emerging Sounds Through Implicit Cooperation: A Novel Model for Dynamic Music Generation”. In Proceedings of the 13th AAAI Conference on Artificial Intelligence and Interactive Digital Entertainment (AIIDE 2017), Utah, USA, October 2017. Download paper Watch video
Avoiding Congestion on Robotic Swarms Navigation
Robotic swarms are decentralized systems formed by a large number of robots. One of the main problems encountered in a swarm is congestion, as a great number of robots often must move towards the same region. This happens when robots have a common target, for example during foraging or waypoint navigation. Hence, we are investigating several novel algorithms to alleviate congestion in such situations. Additionally, we are currently developing new theoretical models to better understand congestion in swarm systems. Our algorithms and models are investigated in realistic simulations, and/or real world experiments.
Paper Examples
- L. S. Marcolino, Y. T. dos Passos, Á. A. F. de Souza, A. S. Rodrigues, L. Chaimowicz. “Avoiding target congestion on the navigation of robotic swarms”. In Autonomous Robots, Vol. 41, No. 6, August 2017, p. 1297-1320, DOI 10.1007/s10514-016-9577-x. Download paper (author version) View Springer Version (SharedIt)
Learning to Keep Cohesion in Swarm Navigation
A robotic swarm is a particular type of Multiagent System that employs a large number of simpler agents in order to cooperatively perform different tasks. Oftentimes, the implementation of complex swarm behaviors is a challenging task, and researchers have started to rely on machine learning techniques, which normally require large and complex training setups. In this project, we explore segregated navigation tasks, in which different groups of robots navigate in shared environments without mixing with others.
Paper Examples
- T. M. Grabe, F. R. Inácio, L. S. Marcolino, D. G. Macharet, L. Chaimowicz. “Stand by me: Learning to keep cohesion in the navigation of heterogeneous swarms”. In the 4th International Symposium on Swarm Behavior and Bio-Inspired Robotics (SWARMS 2021), June 2021. Download paper