2024 ECC workshop:

Vision-based target motion estimation is a fundamental problem that exists in many application domains such as vision-based autonomous driving. We have been studying this problem for several years. Our research was specifically motivated by the interesting aerial target pursuit task. A classic method to solve this problem is the bearing-only approach, where the vision measurement is modeled as a bearing vector. Although this approach has been studied extensively so far, one fundamental limitation of this approach is that it requires the observer to have higher-order maneuvers than the target. This requirement is not practical in many application scenarios. We therefore studied how to improve the observability without the additional maneuvers. In this talk, I will introduce a series of research results that we obtained recently in this field. Related research results have been published in Transactions on Robotics and the International Journal of Robotics Research.

Coverage control is a distributed robot control technology that deploys multiple robots in order to optimize data sampling over a given mission space. Meanwhile, advanced technology for reconstructing 3D scene, typified by Structure-from-Motion (SfM) and visual SLAM, has been highly matured in the field of computer vision. In this talk, we explore the fusion of these two technologies and address issues at their intersection. After briefly reviewing coverage control, we highlight fundamental differences between standard coverage control and image sampling for the 3D model reconstruction. We then reveal that the notion of constraint-based control provides a solution to these novel challenges. It is then demonstrated that the present solution enhances accuracy of the reconstructed 3D model, as compared with the standard coverage control. We moreover exemplify that coverage control with the camera rotational control further improves the model quality. As an application, we then present a crop monitoring problem, and demonstrate the effectiveness of the present technology through field experiments over a vineyard. We finally present our latest work on coverage control with real-time feedback of the 3D model brought by a real-time mapping technology, NeuralRecon. To address the issue, we propose two different solutions, namely fully autonomous control and human-enabled stealthy coverage control.

Visual bearing and range measurements of point landmarks are complimentary measurements. While classical Rigid-body symmetry is the correct symmetry to study point feature landmark measurements, there is a different polar symmetry that is better suited to studying systems with bearing and range measurements. In this talk I introduce the polar symmetry for the bearing and range SLAM problem and show that with respect to this symmetry the measurement linearisation for a SLAM can be significantly improved compared to classical Rigid-body formulation of SLAM.

The determination of spatiotemporal information, i.e., sensor positions and clocks, is crucial for wireless sensor networks to perform coordination and distributed data fusion. Due to its high-precision time-of-arrival measurements in communication and ranging, ultra-wideband (UWB) sensing technology has been widely used for distributed localization and clock synchronization. This talk presents topological conditions on UWB sensor networks necessary to perform localization and/or clock synchronization. By extending graph rigidity theory to the clock and joint position-clock domains, clock rigidity theory and joint rigidity theory are proposed, providing graph-based methods for verifying and constructing networks that can be localized and synchronized. Built on graph analysis, a distributed clock parameter estimator and a distributed position-clock estimator are proposed and supported with convergence guarantees.

We present a solution for locating the source, or maximum, of an unknown scalar field using a swarm of mobile robots. Unlike relying on the traditional gradient information, the swarm determines an ascending direction to approach the source with arbitrary precision. The ascending direction is calculated from field strength measurements at the robot locations and their relative positions concerning the swarm centroid. Rather than focusing on individual robots, we focus the analysis on the density of robots per unit area to guarantee a more resilient swarm, i.e., the functionality remains even if individuals go missing or are misplaced during the mission. We reinforce the algorithm's robustness by providing sufficient conditions for the swarm shape so that the ascending direction is almost parallel to the gradient. The swarm can respond to an unexpected environment by morphing its shape and exploiting the existence of multiple ascending directions. Finally, we validate our approach numerically with hundreds of robots. The fact that a large number of robots with a generic formation always calculate an ascending direction compensates for the potential loss of individuals.

Robots are becoming an integral part of our everyday lives, assisting people with a wide range of tasks. Ensuring verifiably safe robot motion around people and other robots is crucial for achieving truly dependable and reliable autonomous robots. Traditional decoupled planning, control, and perception approaches fall short in ensuring safe autonomous motion for highly dynamic and complex robotic systems. These approaches often consider the geometry of the environment solely for planning, the dynamics of the robot only for control, and the global environmental knowledge for perception. In this talk, I will demonstrate that feedback motion prediction, which involves finding a motion set containing the closed-loop robot motion trajectory, provides a novel safety certificate and interface between planning, control, and perception. This approach systematically relates the geometry of the robot motion to the geometry of the locally sensed environment. I will present analytic feedback motion prediction methods for fully-actuated high-order robot dynamics and nonholonomically constrained unicycle/bicycle dynamics. Additionally, I will showcase applications of feedback motion prediction for sensor-based safe mobile robot navigation around obstacles.

Many studies show that natural systems, such as insects and birds, rely mainly on divergent flow to fly safely near obstacles. By exploring the properties of divergent flow, we have proposed a novel constructive control method that can effectively prevent collisions between neighboring agents while achieving the primary control objective. The proposed method is simpler in terms of design and more efficient in real-time computation than most solutions in the literature. It can also be applied to a broader range of multi-robot systems. It is designed as the sum of a nominal decentralized controller with constructive barrier feedback consisting of divergent flow to slow down the relative velocity towards neighboring agents without affecting the nominal controller's performance. Compared to traditional barrier function-based optimization controllers, the proposed constructive barrier feedback avoids feasibility issues and results in more computationally efficient control algorithms with systematic equilibrium analysis. Applications to the safe platooning and merging control in intelligent transportation scenarios will be illustrated as examples. 

Automated safety-critical control has been a significant challenge for autonomous systems that has received increasing attention in recent years. In this talk we aim to develop automated and computation-efficient methods for autonomous vehicle control and coordination with safety guarantees. We first propose a novel approach to address safety-critical control and provide a universal-formula solution that offers fast computational capabilities tailored for onboard execution in autonomous systems.  We then propose a framework for a multi-vehicle system to fulfill complex tasks represented by temporal logic specifications. Given temporal logic specifications on the coordinated formation and navigation, we develop a controller with runtime safety and convergence guarantees that drive the vehicle group to formally satisfy the specification.