PhD Fellow · Marie Skłodowska-Curie Actions (RAICAM)
Robotics Researcher · FORTH & University of Crete
I design and control modular legged robots — reconfigurable leg modules with learned controllers that adapt to every body they can become.
Modular Legged Robots Robot Design & Control Reinforcement Learning Sim-to-Real
01 · Research Focus
When does a reconfigurable legged robot beat a purpose-built one — and what does that flexibility actually cost?
Reconfigurable Design Context-Conditioned RL Cost-of-Transport Analysis
How can legged robots learn control policies that are fast, stable, and robust enough to run on real hardware?
Reinforcement Learning Whole-Body Control Adversarial Training
What can this robot actually traverse — and how should a planner use that knowledge?
Capability Learning Path Planning Terrain Understanding
Can robots acquire task-oriented manipulation skills by imitating the style — not just the goal — of human motion?
Adversarial Motion Priors Imitation Learning Sim-to-Real
What does it take to move a robot from a simulation result to a machine that works outdoors, in public, on a schedule?
System Integration ROS Field Deployment
02 · Featured Research
An 8-DoF biped module that walks alone or links into a quadruped — built to measure what modularity really costs in reliability, control, and energy.
A low-cost drone and ground vehicle team, built in three weeks and field-tested inside the Zwentendorf nuclear power plant at the European Robotics Hackathon.
A Unitree Go2 walks while keeping a free-rolling ball balanced on a plate mounted on its back — learned in simulation, transferred to the real robot.
03 · Selected Publications
Changda Tian , Panos Trahanias
IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) — Accepted · 2026
Partitioning locomotion rewards by stance and swing roles stabilizes multi-objective reinforcement learning — validated on a real Unitree Go2 and extended to the G1 biped. Accepted at IROS 2026.
Changda Tian , Hamidreza Raei , Arash Ajoudani , Panos Trahanias
2026 IEEE/ASME International Conference on Advanced Intelligent Mechatronics (AIM) · 2026
A quadruped walks while keeping a free-rolling ball balanced on a plate on its back — learned in simulation, transferred to a real Unitree Go2. Presented at IEEE/ASME AIM 2026 in Genova.
Ziqi Ma , Changda Tian , Yue Gao
Robotica · 2025
Human-style, task-oriented manipulation skills learned with adversarial motion priors and deployed on a real Kinova Gen3 arm — now published in Robotica.
Changda Tian , Sasanka Kuruppu Arachchige , Haichuan Li , Juan Jose Garcia Cardenas , Hamidreza Raei , Enes Dincer , Alperen Kenan , Paul Bremner , Manuel Giuliani , Gerhard Neumann , Arash Ajoudani , Adriana Tapus , Tomi Westerlund , Joni-Kristian Kamarainen , Luis Figueredo , Simon Watson , Panos Trahanias
2025 IEEE International Symposium on Safety, Security, and Rescue Robotics (SSRR) · 2025
A low-cost drone + ground-vehicle team for nuclear power plant inspection, built in three weeks with ROS 2 and field-proven at the EnRicH 2025 trials in a real plant — fully open-sourced.
Changda Tian , Yue Gao
2022 IEEE International Conference on Robotics and Biomimetics (ROBIO) · 2022
Best Paper in Biomimetics — Finalist
A hierarchical planner that learns a hexapod's traverse capability across body topologies, enabling reliable long-range locomotion planning.
04 · Timeline
Role-based reward decomposition for legged locomotion reinforcement learning — stance/swing reward partitioning validated on a real Unitree Go2 and extended to the G1 biped.
Sim-to-real reinforcement learning teaches a Unitree Go2 to walk while balancing a ball on its back — presented in Genova.
Started my core PhD project: an 8-DoF biped module that links into a quadruped, and the first measurement of the generality–cost frontier of modular legged robots.
Sim-to-real indoor aerial inspection, presented at the International Conference on Unmanned Aircraft Systems.
Built a low-cost air–ground robot team in three weeks and field-tested it in a real nuclear power plant; published at IEEE SSRR 2025 with a full open-source release.
Human-style manipulation with adversarial motion priors — journal version of the HMAMP system, validated on a real Kinova Gen3 arm.
05 · Open Source
Long-range planning that respects what a legged robot can traverse
Research code for capability-aware locomotion planning on hexapod robots: simulation environments, capability-abstraction training, and capability-augmented path planners.
Python PyTorch CoppeliaSim
Training manipulation policies with human-style motion priors
Training pipeline for adversarial motion priors on robot arms — human video keypoint extraction, robot alignment, and AMP-augmented reinforcement learning, with Kinova Gen3 deployment scripts.
Python PyTorch
A reproducible controller stack for six-legged robots
Layered locomotion control for hexapods: trajectory-optimization gait generation below, learning-based planning above, with ROS integration and sim-to-real tooling.
C++ Python ROS
06 · News
07 · Contact
I'm happy to talk about research, open-source robotics, and collaborations — the fastest way to reach me is email.
