At AIKO, I lead the Virtual Environments department, focusing on real-time simulation for autonomous space missions. The main projects are organized around synthetic data generation for AI-based navigation, reinforcement learning environments for GNC, and interactive VR/AR demos and real-time simulators.
Buzz: RL-controlled In-Orbit Servicing simulator
Buzz is a real-time simulator for AI-based Guidance, Navigation and Control (GNC) in CubeSat rendezvous scenarios.
- Implements a closed-loop environment where an RL agent learns to perform orbital approach and docking between CubeSats.
- Supports training, testing, and validation of the agent under different mission configurations and edge cases.
- Includes a human-vs-RL demo that allows operators to stress-test the agent and showcase the system in an interactive way.
- Integrates orbital and attitude dynamics, optical payload simulation, and hardware subsystems modeling, all built on Unity3D, C#, and Python.
Data generation for visual navigation and pose estimation
Synthetic imagery and telemetry pipelines for visual navigation and pose estimation in proximity operations (e.g., asteroid, CubeSat).
- Generates photorealistic scenarios where real orbital data is scarce or impossible to collect.
- Used to train and validate AI-based pose estimation and navigation pipelines.
- Supports parameterized scenarios to cover edge cases and long-tail distributions.
VR and interactive demos for space AI
Immersive experiences and interactive tools to make space technologies accessible and understandable.
- VR demo for an AI-based lunar landing agent, showcasing how reinforcement learning can be applied to complex GNC tasks.
- Interactive UI for controlling and debugging simulation scenarios in real-time, with visual feedback on dynamics, sensors, and agent behavior.
- Used for product demos, outreach, and internal training.
Earth Observation payload generator
Synthetic data generator for Earth Observation (EO) payloads, supporting algorithm development and validation where real acquisitions are limited.
- Simulates EO imagery and telemetry under various mission configurations.
- Used to train and test ML models for scene analysis, object detection, and change detection.
- Helps bridge the gap between research prototypes and deployable systems.
Flight Dynamics Engine
Core component of the AIKO simulation ecosystem, providing orbital and attitude dynamics, sensor models, and actuator models for GNC and mission analysis.
- Implements numerical methods for accurate orbital mechanics and attitude evolution.
- Integrates with Unity-based simulators and RL environments.
- Serves as the backbone for multiple projects, from Buzz to EO data generation.