Malintha Fernando

I am a Digital Futures Postdoctoral Fellow at the KTH Royal Institute of Technlogy, Sweden. My research lies at the intersection of multi-agent systems, game theory and reinforcement learning.

My current research revolves around the following questions: (1) how can we effciently learn distributed RL policies to make decisions, under limited budget and profit stochasticity in routing settings (i.e., vehicle routing, prize-collection), and (2) How can we leverage tools from game theory to handle situations involving competing interests? (3) How can we best allocate profits among competing parties to improve the social-optimality (get as clost to the optimal solution, i.e, minimize the prize-of-anarchy?)

Answering these questions will enable autonomous fleet managers make optimal, sequential, decisions in a distributed manner under both competitive, and cooperative settings. E.g., given the observations about the competitors, and the own fleet, which depot should each vehicle cover?

From a policy-making perspective it will allow predicting which areas will be underserved by the service providers in pursuit of profit. How much should we tax/incentivize the providers for fairer service distribution?

My current research is a natual extension of my Ph.D., in which I researched coordinating drone swarms for parcel delivery, formation control, under partial observations. I am advised by Prof. Silun Zhang (KTH Mathematics), and Prof. Petter Ögren (EECS). Before joining KTH, I briefly worked as a machine learning lecturer at Indiana University, Bloomington, USA, where I obtained my Ph.D.

I am currently in the job market, I am open for both academic and industry positions in deep learning, decision-making, autonomous systems. Shoot me a mail, if you think I am a fit.

News

Mar, 2026	We organized the Digital Futures Young Scientists Conference, 2026. It was a fantastic day at Stockholm with some very intersting Keynotes. Read the newstory published on the DF official website here.
Feb, 2026	Can counterfactuals speed up learning RL policies to solve decision-making problems on graphs? We cetainly think so. A paper is in the making.
Nov, 2025	Presented our work “Competing Collaborators: Strategic Planning on Graphs” at “Digitalize i Stockholm 2025”
Oct, 2025	A preprint is out. We propose the concept of “Ordinal-Rank Conditioning” to improve reinforcement learning agents making-decisions on graphs under rank-based conflict resolution rules. Arxiv
Jul, 2025	I have spent the past few months studying the problem which we refer to as the “Stochastic Prize-Collecting Games”. We have obtained insteresting results with a fictitious-play-based learning mechanism in the presence of rank-based agents. Await for our preprint.

Selected Research Projects

For an updated list, please visit my google scholar page.

Graph Attention Aerial Fleet Coordination

This work propose a multi-step partially-observable stochastic game formulation for coordinating fully-autonomous heterogeneous eVTOL fleet with graph-attention multi-agent reinforcement-learning (MARL). The paper got accepted to Robotics: Science and Systems, 2023 . [Paper].

Video link to my presentation at RSS 2023.

Graphical Game-Theoretic Communication-Aware Coverage

Wireless communication plays a crucial role in distributed coordination control of robot swarms. In this work we propose a robust, real-time coverage control approach for UAV swarms to provide the wireless coverage for a ground robot team using a UAV team operating over large geographic region with the local communication. The two images show a static and mobile ground robot team, where the UAV fleet is changing their formation to maximize the coverage for the ground robots. The dynamic communication links among the UAVs are shown in grey.
Checkout the full demo on Youtube .

Two static (Left) and moving (Right) ground robot fleets.

Real-Time Distributed Flocking Control

Murmurations are one of the most beatiful natural phenomena in the world, and characterizes the perfect swarm . In this work I tried to recreate the flocking characteristics UAVs using a variational inference approach in a distributed manner. The drones only uses their local observations to infer the control actions from feasible set; which makes this approach unique compared to the literature which we can guarantee the covergence and the dynamical feasibility.
Checkout the full demo on Youtube .

Left:A real robot demonstration using Crazyflie nano-drone team. Right: Two Carzyflie teams merging in simulation.

Multi-UAV Formation Control

Drone formation control has been gaining a lot of attention lately with applications ranging from entertainment to and defense industries. In this work I propose a rigid-body-based formation controlling approach with drone aggressive trajectory generation. The drones showed the ability to move at 2.5ms^-1 (over 10 body-lengths a second) in a user defined formation with real-time receding horizon trajectory planning.
Checkout the formation control demo on Youtube .
Checkout the minimum-jerk trajectory generation demo on Youtube .

Left: Drone team of 5 moving in pentagon formation in a small indoor environment. Right: A SloMO of a Single UAV moving with precise aggressive motion through traffic cones in a minimum-jerk trajectory.

Selected Publications

Graph Attention Multi-Agent Fleet Autonomy for Advanced Air Mobility

Malintha Fernando, Ransalu Senanayake, Heeyoul Choi, and 1 more author

2023
CoCo Games: Graphical Game-Theoretic Swarm Control for Communication-Aware Coverage

Malintha Fernando, Ransalu Senanayake, and Martin Swany

IEEE Robotics and Automation Letters, 2022
Online flocking control of UAVs with mean-field approximation

Malintha Fernando

In 2021 IEEE International Conference on Robotics and Automation (ICRA), 2021
Formation control and navigation of a quadrotor swarm

Malintha Fernando, and Lantao Liu

In 2019 International Conference on Unmanned Aircraft Systems (ICUAS), 2019