Pranav Rajbhandari

Machine Learning Department

Carnegie Mellon University

Pittsburgh, USA

I am a PhD student at CMU’s Machine Learning Department, advised by Prof. Tuomas Sandholm. My current research is focused on developing algorithms to solve large-scale imperfect information games.

My general interests are AI/ML, Topology, and long walks on the beach.

My general interests lie in the disjoint union of developing Machine Learning algorithms for strategic scenarios, and Topology/discrete mathematics.

Research statement here.

Inspired by my friend's website, I have also included reflections on my coursework: Coursework Reflection

news

Jul 19, 2025	Making a website

selected publications

arXiv

Distributed Constraint Optimization via Online Learning and Iterative Pricing with Application to Large-Scale Satellite Scheduling

Itai Zilberstein, Pranav Rajbhandari, Steve Chien, and Tuomas Sandholm

2026

Abs PDF Code

Distributed constraint optimization problems (DCOPs) provide a popular framework for distributed decision making under limited communication, but many real-world instances are too large to solve monolithically. We address this challenge from two complementary directions. We revisit the connection between DCOPs and potential games, and adapt modern online learning algorithms for equilibrium finding to DCOPs. We show that these algorithms are competitive with representative incomplete DCOP algorithms. We then turn to decomposition frameworks for large-scale DCOPs, motivated by large-scale decentralized satellite scheduling. We propose a new framework that separates a DCOP into two interacting subproblems: a high-level meta-DCOP for task allocation, and independent local optimization problems for scheduling. To couple the two levels, we develop a novel iterative pricing method that updates the meta-level utilities using feedback from the local optimizers. Combining our online learning methods with our iterative pricing framework, we obtain near-optimal performance on real-world decentralized satellite scheduling problem instances, fulfilling over 99% of observation requests compared with 87% for state-of-the-art baselines.
Combinatorics at the Confluence

Geodesic complexity of the octahedron

Florian Frick and Pranav Rajbhandari

2026

Abs PDF Poster Code

The geodesic complexity of a length space $X$ quantifies the required number of case distinctions to continuously choose a shortest path connecting any given start and end point. We prove a local lower bound for the geodesic complexity of $X$ obtained by embedding simplices into $X\times X$. We use this technique, along with an algorithm to compute the cut locus on polyhedra surfaces, to prove the geodesic complexity of the octahedron is four. Our method is inspired by earlier work of Recio-Mitter and Davis, and thus recovers their results on the geodesic complexity of the $n$-torus and the tetrahedron, respectively.
arXiv

Geodesic complexity of the octahedron, and an algorithm for cut loci on convex polyhedra

Florian Frick and Pranav Rajbhandari

2025

Abs PDF Poster Code

The geodesic complexity of a length space $X$ quantifies the required number of case distinctions to continuously choose a shortest path connecting any given start and end point. We prove a local lower bound for the geodesic complexity of $X$ obtained by embedding simplices into $X\times X$. We additionally create and prove correctness of an algorithm to find cut loci on surfaces of convex polyhedra, as the structure of a space's cut loci is related to its geodesic complexity. We use these techniques to prove the geodesic complexity of the octahedron is four. Our method is inspired by earlier work of Recio-Mitter and Davis, and thus recovers their results on the geodesic complexity of the $n$-torus and the tetrahedron, respectively.
arXiv

Understanding visual attention beehind bee-inspired UAV navigation

Pranav Rajbhandari, Abhi Veda, and 3 more authors

2025

Abs PDF Poster

Bio-inspired design is often used in autonomous UAV navigation due to the capacity of biological systems for flight and obstacle avoidance despite limited sensory and computational capabilities. In particular, honeybees mainly use the sensory input of optic flow, the apparent motion of objects in their visual field, to navigate cluttered environments. In our work, we train a Reinforcement Learning agent to navigate a tunnel with obstacles using only optic flow as sensory input. We inspect the attention patterns of trained agents to determine the regions of optic flow on which they primarily base their motor decisions. We find that agents trained in this way pay most attention to regions of discontinuity in optic flow, as well as regions with large optic flow magnitude. The trained agents appear to navigate a cluttered tunnel by avoiding the obstacles that produce large optic flow, while maintaining a centered position in their environment, which resembles the behavior seen in flying insects. This pattern persists across independently trained agents, which suggests that this could be a good strategy for developing a simple explicit control law for physical UAVs.
AAMAS

Transformer Guided Coevolution: Improved Team Formation in Multiagent Adversarial Games

Pranav Rajbhandari, Prithviraj Dasgupta, and Donald Sofge

In Proc. of the 24th International Conference on Autonomous Agents and Multiagent Systems (AAMAS 2025), Detroit, Michigan, USA, 2025

Abs PDF Poster Code

With the increasing number of autonomous platforms in everyday life, forming coordinated teams of agents becomes vital. To solve this, we propose BERTeam, an algorithm inspired by Natural Language Processing. BERTeam trains a transformer-based deep neural network to select from a population of agents. It can integrate with coevolutionary deep reinforcement learning, which evolves a diverse set of players to choose from. We evaluate BERTeam in Marine Capture-The-Flag, and find it learns non-trivial team compositions that outperform unknown opponents. In this setting, we find that BERTeam outperforms MCAA, another team selection algorithm.
IEEE ROBIO

Learning NEAT Emergent Behaviors in Robot Swarms

Pranav Rajbhandari and Donald Sofge

In 2024 IEEE International Conference on Robotics and Biomimetics (ROBIO), 2024

Abs PDF DOI

When researching robot swarms, many studies observe complex group behavior emerging from the individual agents' simple local actions. However, the task of learning an individual policy to produce a desired group behavior remains a challenging problem. We present a method of training distributed robotic swarm algorithms to produce emergent behavior. Inspired by the biological evolution of emergent behavior in animals, we use an evolutionary algorithm to train a 'population' of individual behaviors to produce a desired group behavior. We perform experiments using simulations of the Georgia Tech Miniature Autonomous Blimps (GT-MABs) aerial robotics platforms conducted in the CoppeliaSim simulator. Additionally, we test on simulations of Anki Vector robots to display our algorithm's effectiveness on various modes of actuation. We evaluate our algorithm on various tasks where a somewhat complex group behavior is required for success. These tasks include an Area Coverage task and a Wall Climb task. We compare behaviors evolved using our algorithm against designed policies, which we create in order to exhibit the emergent behaviors we desire.