This presentation describes my academic and research trajectory, addressing three main topics: (i) active perception, (ii) robust trajectory generation, and (iii) human-robot interaction. In particular, the aim is to demonstrate how principled optimization techniques can address critical challenges in modern robotics while maintaining computational efficiency for real-time applications.

The research on active perception focuses on finding the optimal inputs for a mobile robot in order to minimize estimation uncertainty. By integrating information-theoretic (based on the Constructibility Gramian) objectives into trajectory planning algorithms, frameworks are developed that enable robots to autonomously prioritize configurations for optimal perception performance. The proposed methodologies demonstrate how trajectory optimization directly enhances state estimation in autonomous systems.

Regarding robust trajectory generation and planning, sensitivity-aware methods address parameter uncertainty through closed-loop sensitivity analysis. Rather than relying on conservative worst-case scenarios, the proposed approaches generate trajectories inherently robust to parameter variations. Two complementary strategies are presented: a global motion planner that leverages machine-learning techniques for efficient sensitivity prediction, and a sensitivity-aware Model Predictive Control (MPC) framework that embeds robustness guarantees directly into real-time feedback control.

The research on human-robot interaction focuses on physical collaboration between humans and robotic systems. An MPC framework for the collaborative transportation of an object between a human and a drone is presented. This research direction also proposes machine-learning techniques to predict human-related information (e.g., human fatigue and ergonomic risks), allowing the robot to adapt its autonomy in real-time as a function of the human operator's behavior.

Experimental validation across diverse platforms (from quadrotors to ground manipulators) demonstrates the practical applicability of the proposed methods through implementation on real robotic systems.

The presentation concludes with future perspectives that frame the necessity of having stability guarantees for neural networks and illustrates the objective to switch from a single-agent to a heterogeneous multi-robot system as a natural evolution of the previously addressed challenges. This envisioned application represents the convergence of the three research pillars, where the necessity for distributed sensing drives the need for active perception, the presence of multiple sources of uncertainty (e.g., inter-robot parameters) necessitates robust, sensitivity-aware planning, and the need for safe collaboration frameworks for advances in human-robot interaction and decision-making. This evolution introduces the additional "multi-robot challenge", managing the trade-off between decentralized autonomy and global system coherence. To address this, a hierarchical architecture that includes planning, machine-learning, and control is proposed to enable heterogeneous teams to operate safely and effectively in complex, dynamic environments.