Miroljub Mihailovic

Abstract — Developed a hybrid control framework for lower-limb exoskeletons integrating vision-based environment perception and kernelized movement primitives (KMPs) for adaptive gait generation. The method learns a human gait patterns both in task and joint space, applying real-time linear constraints based on RGB-D perception to ensure smooth, safe, and environment-aware locomotion across multiple terrains (slopes, stairs, obstacles). Validated on a commercial exoskeleton with both simulated and real-world experiments.

Abstract — Developed a modular control framework for adaptive exoskeleton locomotion integrating robotic vision, neural predictive control, and STL-based planning. The system produces coordinated multistep gait trajectories in real time, enabling smooth and adaptive motion over complex terrain.

Abstract — Developed an egocentric vision framework combining exoskeleton-mounted and AR visor cameras to build a unified 3D map for adaptive gait planning and real-time obstacle avoidance. The system integrates perception, planning, and augmented reality feedback for safer and more intuitive locomotion.

Abstract — Experimental validation of an autonomous rendezvous and docking framework for CubeSat mock-ups in microgravity. The developed system integrates sensor fusion, relative navigation, and real-time control to execute fully autonomous approach and docking manoeuvres during parabolic flight campaigns. Achieved centimetre-level alignment accuracy and high docking reliability under realistic space-like conditions.

Abstract — The ERMES project investigates autonomous docking strategies for nano satellites under microgravity conditions. A hardware-in-the-loop setup was deployed aboard parabolic flights to assess docking dynamics between active and passive CubeSat mock-ups. The study provides insights into guidance, navigation, and control limitations in low-gravity environments and establishes a foundation for future autonomous proximity operations in small satellite missions.

My focus lies in two critical areas: executing unit tests and performing code refactoring for ROS-Neuro. These concerted efforts are aimed at bolstering reliability, validating functionality, and optimizing the codebase for enhanced maintenance and scalability.

Designed and implemented the CubeSat’s Onboard Computer System, control logic, and sensor fusion algorithms for real-time autonomous docking under microgravity conditions.

Developed a reinforcement learning policy integrated into a shared teleoperation system. The agent navigates unknown environments with obstacle avoidance, compared to traditional rule-based policies for safety and efficiency.