Agon Serifi

I am an Associate Research Scientist with the Disney Research Robotics team. My work focuses on deep reinforcement learning and human motion to develop control methods that enable robots to generalize and exhibit rich, diverse behaviors. This includes motion imitation, generation, and the reproduction of lifelike behavior. I received my PhD in Computer Science (2025) from the Computer Graphics Laboratory (CGL) at ETH Zurich under Prof. Markus Gross and Dr. Moritz Bächer, in collaboration with Disney Research. I also hold a Bachelor's (2019) and a Master's (2021) degree in Computer Science from ETH Zurich.

News

Loading news

Highlight

Research

Loading research papers

Kamino: GPU-based Massively Parallel Simulation of Multi-Body Systems with Challenging Topologies

Vassilios Tsounis, Guirec Maloisel, Christian Schumacher, Ruben Grandia, Agon Serifi, David Müller, Chris Amevor, Tobias Widmer, Moritz Bächer

arXiv 2026

project page | paper | code | ArXiv

This paper presents Kamino, a GPU-based physics solver for massively parallel simulation of heterogeneous, highly-coupled mechanical systems. Built in Python with NVIDIA Warp and integrated into the Newton framework, Kamino natively supports complex topologies with kinematic loops, avoiding tree-based approximations. It enables batched simulation of structurally diverse robots on a single GPU and demonstrates its capabilities through reinforcement learning policy training on challenging robotic systems.

Olaf: Bringing an Animated Character to Life in the Physical World

David Müller*, Espen Knoop*, Dario Mylonopoulos, Agon Serifi, Michael A. Hopkins, Ruben Grandia, Moritz Bächer

IEEE Robotics and Automation Letters (RA-L) 2026

paper | video | IEEE

In this work, we bring Olaf, an animated character, to life in the physical world, relying on reinforcement learning guided by animation references for control.

Robot Crash Course: Learning Soft and Stylized Falling

Pascal Strauch, David Müller, Sammy Christen, Agon Serifi, Ruben Grandia, Espen Knoop, Moritz Bächer

ICRA 2026

paper | video | ArXiv

This paper presents a method for learning controlled, soft falls for bipedal robots. Rather than preventing falls, we focus on minimizing physical damage and achieving desired end poses through a robot-agnostic reward function that balances impact minimization with pose control during reinforcement learning. The approach uses simulation-based sampling to handle diverse falling conditions and enables arbitrary end pose specification at inference time, demonstrating robust falling behaviors in both simulation and real-world experiments.

AMOR: Adaptive Character Control through Multi-Objective Reinforcement Learning

Lucas N. Alegre*, Agon Serifi*, Ruben Grandia, David Müller, Espen Knoop, Moritz Bächer

SIGGRAPH 2025

project page | paper | video | ACM

This paper presents AMOR, a method for adaptive character control through multi-objective reinforcement learning. A single policy is trained conditioned on reward weights and motion context, enabling post-training tuning of reward trade-offs and zero-shot adaptation to diverse tracking behaviors. The system allows for sim-to-real transfer of dynamic motions and robust control across a broad range of tasks, with optional hierarchical extensions for task-driven weight modulation.

Autonomous Human-Robot Interaction via Operator Imitation

Sammy Christen, David Müller, Agon Serifi, Ruben Grandia, Georg Wiedebach, Michael A. Hopkins, Espen Knoop, Moritz Bächer

IROS 2025

project page | paper | video | IEEE

This paper presents a method for autonomously controlling expressive human-robot interactions by training a diffusion-based imitation model on real operator data. The resulting system predicts both continuous and discrete control inputs while conditioning on human pose, allowing the robot to produce robust, lifelike interactions and diverse moods that closely resemble expert operator performance in real-world scenarios.

Robot Motion Diffusion Model: Motion Generation for Robotic Characters

Agon Serifi, Ruben Grandia, Espen Knoop, Markus Gross, Moritz Bächer

SIGGRAPH Asia 2024

project page | paper | video | ACM

This paper presents a method for aligning text-conditioned kinematic motion generators with the capabilities of robotic characters. Our approach combines motion generation with physics-based character control. First, we train a critic that acts as a reward surrogate by predicting the performance of the downstream non-differentiable control task. The critic provides an efficient and differentiable loss function, which is subsequently used to align the motion generator more closely with the character's physical capabilities.

Spline-Based Transformers

Prashanth Chandran*, Agon Serifi*, Markus Gross, Moritz Bächer

ECCV 2024 (Oral)

project page | paper | supp | video | Springer

A new class of Transformer models that eliminate the need for positional encoding. Spline-based Transformers embed an input sequence of elements as a smooth trajectory in latent space. Fully leveraging the continuity inherent in sequential data, this approach significantly outperforms traditional transformer models in efficiency and accuracy.

VMP: Versatile Motion Priors for Robustly Tracking Motion on Physical Characters

Agon Serifi, Ruben Grandia, Espen Knoop, Markus Gross, Moritz Bächer

SIGGRAPH/Eurographics Symposium on Computer Animation (SCA) 2024

project page | paper | video | ACM

This paper introduces a two-stage framework for transferring expressive and dynamic motions onto robotic characters. Starting from a large corpus of human motion capture data, we first learn a latent space representation of short motion windows. This latent space then serves as a prior for training a tracking controller through reinforcement learning. The resulting controller pushes the boundaries of robotic characters, enabling them to move like humans and perform lifelike, dynamic motions such as dancing.

Transformer-Based Neural Augmentation of Robot Simulation Representations

Agon Serifi, Espen Knoop, Christian Schumacher, Naveen Kumar, Markus Gross, Moritz Bächer

IEEE Robotics and Automation Letters (RA-L), presented at IROS 2023

project page | paper | video | IEEE

This paper introduces a modular approach for neural augmentation of simulation states, improving robot simulation accuracy by targeting individual building blocks such as actuators and rigid bodies separately.