Dabin Kim

This paper presents a real-time safety filter for robot navigation that maintains visual feature visibility by minimally adjusting velocity commands, ensuring reliable pose estimation even in GPS-denied environments. Validated in both simulation and real-world SLAM scenarios, the method outperforms standard controllers by preserving high-quality localization.

This survey reviews recent advances in safety-critical motion planning and control for mobile robots, focusing on receding horizon methods and safety filtering approaches. It highlights key challenges in dynamic environments and outlines future directions for developing efficient, formally safe control systems.

This work proposes a novel method for computing constraint admissible positively invariant (CAPI) sets for general reference tracking using neural Lyapunov functions with piecewise-affine activations. By reformulating the problem into linear programs and introducing a learning-based estimator, the approach enables real-time applicability and is validated through simulations and integration with an explicit reference governor.

This paper presents a novel reference governor for multirotor control that enforces visibility constraints while tracking time-varying references, ensuring safe vision-based navigation. The method guarantees theoretical feasibility and is validated through both simulations and real-world experiments.

This letter introduces an efficient online multi-agent trajectory planning algorithm that ensures safe, dynamically feasible paths using a linear safe corridor without relying on soft constraints. Validated through simulations and real-world quadrotor flights, the method achieves fast, deadlock-free planning for up to 60 agents in complex environments.

This paper proposes a perception-aware path planner for MAVs that leverages topological information to select visually rich routes, enhancing visual navigation accuracy and efficiency. By analyzing topological classes derived from Voronoi diagrams, the method outperforms sampling-based planners in both simulation and real-world tests.

This letter proposes a distributed multi-robot exploration algorithm that enables real-time mapping and peak-seeking in unknown environments using Gaussian process regression. The approach supports online learning, decentralized coordination, and collision avoidance, and is validated through simulations and real-world UAV experiments.

This article introduces the T³-multirotor, a novel fully actuated multirotor platform capable of independent six-degree-of-freedom flight through a unique tilting-thruster mechanism. A robust control algorithm is developed and validated in both simulations and experiments, showcasing the platform’s ability to surpass conventional multirotor limitations.

We present a multirotor-based aerial manipulator capable of interacting with a moving structure—a hinged door—using model predictive control and robust trajectory tracking. The proposed system enables real-time, collision-aware manipulation of constrained environments, validated through hardware experiments.