Projects

Kajima: AI-Guided Aerial Risk Screening for Coastal Decision Support

Principal Investigator(s): 
Ricardo Sanfelice (UCSC)
Executive Summary: 

The primary motivation of this project stems from the increasing risks in coastal zones driven by climate change, erosion, and urban pressure. Early identification of surface-level hazards is crucial for environmental monitoring, disaster prevention (such as landslides or tsunamis), ecological research, and infrastructure protection. To address this, the project aims to model, predict, and control complex behaviors by developing an intelligent drone-based system that autonomously screens risk areas near ground surfaces. Using satellite imagery as a prior, AI algorithms will direct drone deployment toward locations of interest, capturing and analyzing continuous trajectory data to evaluate potential hazards like unstable terrain, illegal construction, or ecological damage. For this initial preliminary effort, the goal is to architect a framework for direct data processing software pipeline capable of obtaining trajectory data of physical events. In a longer follow-up effort, this framework will be applied to model, predict, and control complex behaviors across domains such as coastal changes, car/traffic behavior, and animal migrations.

Alianza MX: From Fields to Tables: Robotic Solutions for Sustainable Berry Harvesting

Principal Investigator(s): 
Ricardo Sanfelice (UCSC), Co-PIs: Stefano Carpin (UC Merced), Manuel Navarro-Gutiérrez (Tecnológico de Monterrey), Jorge Isaac Chairez Oria (Tecnológico de Monterrey)
Executive Summary: 

Motivated by the need to address agricultural labor shortages and inefficiencies, this project aims to develop robotic solutions for efficient and sustainable berry harvesting in greenhouses in Mexico and California. By utilizing robots equipped with sophisticated sensors and algorithms, this project intends to automate the berry picking process, ensuring precise identification and collection of ripe berries, thereby enhancing efficiency, reducing waste, and improving overall profitability for farmers.

Data Fusion of GNSS with LEO Satellites as Alternate-PNT Systems Including Anomaly Detection

Principal Investigator(s): 
Ricardo Sanfelice (UCSC)
Executive Summary: 

The objective of this proposed project is a novel software-defined, resilient, network-based Position, Navigation, and Timing (PNT) system that has the potential to provide robust PNT for single-weapon and network-enabled weapon systems of interest to AFRL/RWWN. This proposed technology efficiently utilizes network signals from Low Earth Orbit (LEO) satellites (LEO PNT), Time Difference Of Arrival (TDOA), and Kalman filtering methods to provide accurate and precise PNT. Note that there is no need to collaborate with, or communicate via, the LEO satellites themselves. Rather, the proposed technology uses the LEO Satellites as RF Signals Of Opportunity (SOOP). Reference stations (stationary or mobile) at known locations measure TOA data of the LEO satellites that they can opportunistically receive. These TOA data are securely streamed to Command-and-Control (C&C) servers located in-theatre or secured in the cloud. The C&C servers compute position, velocity, altitude (PVA) for all of the LEO satellites that are in-theatre and stream the resulting LEO PVA and TOA to the in-theatre assets e.g., Aircraft, Drones, Munitions, and Mounted or Dismounted warfighters. Given the PVA and TOA, this technology enables assets to navigate to waypoints and targets to sub-meter accuracy using alt-PNT as a robust alternative to the easily spoofable and jammable Global Navigation Satellite Systems (GNSS). Proposed alt-PNT algorithms are modular to facilitate integration with existing RF network interfaces or use advanced low-probability of detection/interference (LPOD/I) Ultra-Wide-Band (UWB) network transceivers, and very importantly, their implementation is intended on COTS Software Defined Radios (SDR). All PVA and TOA data are recorded at the C&C servers for forensic analysis and algorithm testing and improvement.

MultiHyRL: Robust Hybrid RL for Obstacle Avoidance against Adversarial Attacks on the Observation Space

Principal Investigator(s): 
Ricardo Sanfelice (UCSC), Prashant Ganesh (EpiSci)
Executive Summary: 

The objective of this work is to generate new fundamental science for hybrid dynamical systems that enables systematic design of algorithms using reinforcement learning (RL) tech- niques that are hybrid. Hybrid systems are dynamical systems with intertwined continuous and discrete behavior. Hybrid controllers are algorithms that involve logic variables, timers, memory states, along with the associated decision-making logic. The combination of such mixed behavior, both in the system to control and in the algorithms, is embodied in key future engineering systems. The future autonomous systems will have variables that change continuously according to physics laws, exhibit jumps due to controlled switches, replanning of maneuvers, on-the-fly redesign, and failures, while the control algorithms require logic to adapt to such abrupt changes. Hybrid behavior also emerges in such engineering systems due to their complexity. In fact, communication events, abrupt changes in connectivity, and the cyber-physical interaction between vehicles, humans, robots, their environment, and com- munication networks lead to impulsive behavior that interacts with physics and computing. In this project, we propose to develop novel hybrid reinforcement learning control algorithms that can be validated in experimental data-driven testbeds. The proposed combination of hybrid control and learning informed by real-time data exploits – in a holistic manner – key robust stabilization capabilities of hybrid feedback control and learning capabilities of RL for autonomous systems of interest to AFRL that are part of the broad mission of the DoD.

Learning Algorithms for Hybrid Dynamical Systems using Experimental Data

Principal Investigator(s): 
Prof. Ricardo Sanfelice (UCSC)
Executive Summary: 

The objective of this work is to generate new fundamental science for hybrid dynamical systems that enables systematic design of algorithms for learning the parameters, dynamics, and constraints of such complex systems, using experimental data. Hybrid systems are dynamical systems with intertwined continuous and discrete behavior. Hybrid controllers are algorithms that involve logic variables, timers, memory states, along with the associated decision-making logic. The combination of such mixed behavior, both in the system to control and in the algorithms, is embodied in key future engineering systems. The future autonomous systems will have variables that change continuously according to physics laws, exhibit jumps due to controlled switches, replanning of maneuvers, on-the-fly redesign, and failures, while the control algorithms require logic to adapt to such abrupt changes. Hybrid behavior also emerges in such engineering systems due to their complexity. In fact, communication events, abrupt changes in connectivity, and the cyber-physical interaction between vehicles, humans, robots, their environment, and communication networks lead to impulsive behavior that interacts with physics and computing. The intellectual impact of the proposed research plan stems from a novel use of hybrid control theory, one that leads to experimental data-driven learning algorithms for hybrid systems that are not only robust but also optimal. The proposed combination of hybrid control and optimization in- formed by real-time data exploits–in a holistic manner–key robust stabilization capabilities of hybrid feedback control and optimality guarantees of receding horizon control for com- plex autonomous systems of interest to AFOSR that are part of the broad mission of the DoD.

AFOSR: Verification and Validation of Autonomous Systems with Hybrid Dynamics under Uncertainty

Principal Investigator(s): 
Ricardo Sanfelice (UCSC), Michael Wehner (U. Wisconsin), Daniele Venturi (UC Santa Cruz), Shai Revzen (U. Michigan)
Executive Summary: 

In conventional design processes, the design of the plant, the controller, the prototype as well as the certification of validity are products of consecutive phases of development utilizing distinct simulation, fabrication, and synthesis tools. Each phase produces an “optimal” solution, which is typically not jointly optimal for all phases. Phases need to be restarted from scratch if a new edge case or catastrophic failure is discovered at a later stage, and the interaction between failures, fabrication tolerances, and model inaccuracy is murky at best. Instead, we propose to combine rapid prototyping, hybrid systems techniques for modeling and control, formal verification, and quantified uncertainty and risk models for systematic autonomous system development. New advanced systems have raced ahead of our ability to analyze them, while advanced manufacturing technologies allow us to quickly and inexpensively build them. The paradigm we propose rapidly prototypes the design, validates its fabrication, and quantifies both its performance and its failure risk by conducting physical tests of the prototypes. This allows the new paradigm to rationally improve this combined representation without being subject to reality-gaps in simulation, and without requiring catastrophic restarts when new edge cases and failure modes are discovered.

NSF: Collaborative Research: CPS: Medium: Constraint Aware Planning and Control for Cyber-Physical Systems

Principal Investigator(s): 
Prof. Ricardo Sanfelice (UCSC), Prof. Shai Revzen (UMich)
Executive Summary: 

The objective of this work is to generate new fundamental science for computer controlled complex physical systems, a broad class of cyber-physical systems (CPS), and demonstrate this science in aerial vehicles and walking robots.  The new science enables autonomous planning and control in the presence of failures and abrupt changes in system variables.  A framework for the design of algorithms that exploit awareness of the physical and design constraints to autonomously self-adapt their motion plan and control actions will be generated.  The approach exploits elements from geometry, adaptive control, and hybrid control to advance the knowledge on modeling, planning, and design of CPS with constraints, nonsmooth, and intertwined continuous and discrete dynamics.  Unlike current approaches, which separate the task associated with planning the motion from the design of the algorithm used for control, the algorithms to emerge from this project self-learn and self-adapt in real time to cope with unexpected changes in motion and specification constraints so as to enable autonomous systems to perform robustly and safely, and degrade gracefully under failure conditions.  Specifically, the new algorithms will learn and monitor the physical and design constraints in real time, and adapt both planner and controller by selecting the appropriate constraints to enforce, with robustness and safety guarantees.  The capabilities of the new tools will be demonstrated on multi-legged robots in harsh environments that make them prone to failures, and on aerial vehicles in contested/adversarial environments.

The proposed plan contributes to Science of Cyber-Physical Systems by addressing modeling, motion planning, and design of CPS with constraints, nonsmooth, and intertwined continuous and discrete dynamics. The merits of the proposal fall into four broad categories: (i) a framework to mathematically formulate learning-based planning and control for CPS with awareness of its constraints, (ii) novel architectures that lead to robust adaptive constraint satisfaction,  (iii) deep understanding of roles and priorities of system constraints in CPS, and (iv) tools and design techniques that permit engineers to deploy constraint aware algorithms. The results of this work are broad in their application to all kinds of CPS that require planning and control, in particular, autonomous systems in transportation (air and ground). Synergistic collaborations with researchers at Samsung, the start-up Ghost Robotics, and at the University of Bologna are instrumental in disseminating the application of our results to industry and academia.  A synergistic outreach program at UCSC and the University of Michigan impacts high school students and teachers.

AFOSR SURI: Digital Twin Enabled Autonomous Control for On-Orbit Spacecraft Servicing

Principal Investigator(s): 
Prof. R. Sanfelice (UCSC), Prof. I. Kolmanovsky (U. Mich), Prof. K. Willcox (U. Texas at Austin), Prof. D. Venturi (UCSC)
Executive Summary: 

This multidisciplinary program of basic and applied research addresses the challenges of autonomous on-orbit space operations. A novel program of mathematical, computational and engineering research addresses foundational challenges in 1) digital twin technology, 2) feedback control algorithms and 3) human-machine-aware supervisory schemes, all in the context of space systems. A campaign of experimental validation and demonstration in multiple testbeds serves to integrate the research advances; validate, robustify and harden the algorithms; and enable transition of research products into DoD applications. Key novel research ideas include: 1) a probabilistic graphical model foundation for digital twins that automates model calibration and updating, makes data-model integration scalable across many assets and quantifies uncertainty; 2) new feedback control algorithms that leverage model predictive control to achieve control for on-orbit space operations under stringent constraints, and handle uncertainty by accounting for hybrid and set-valued dynamics; 3) new human-machine-aware supervisory schemes targeting uncertain and contested operating environments, synthesized as hybrid control schemes that coordinate control algorithms across different phases of the mission, address recovery from unexpected conditions, and incorporate the role of potential human intervention. Experimental demonstration in testbeds covers mission-specific features in three case studies: on-orbit assembly, decommissioning and towing, and on-orbit refueling. Experiments will be conducted at ERAU, UCSC, NPS, and AFRL/RV, specifically, in the multiple testbeds at the ROC lab, the BONSAI lab, the SPACER lab, and the LINCS lab. The experimental phase is a stepping stone for the development of a flight demonstration in collaboration with AFRL/RV and industry partners. The effort involves industry partners with unique expertise in space operations and the case studies driving the project.

STMicroelectronics: Incorporating STMicroelectronics Drone Kit in UC Santa Cruz’s Robotics Engineering Program

Principal Investigator(s): 
Prof. Ricardo Sanfelice (UCSC)
Executive Summary: 

Developing high-quality drone technology is essential for the emerging global market, as drones are able to perform increasingly complex tasks with limited human intervention. Businesses and defense organizations are increasing their investments in drone research to aid in military emergency response, disaster relief, and damage assessment during wildfires, hurricanes, and earthquakes. Utilizing innovative engineering technology to design high quality, autonomous drones is our goal.

As building drones helps students develop valuable skills in robotics and hardware design, we propose a project to incorporate STMicroelectronics drone kit STEVAL-DRONE01 in our educational program. The goal of the project is to design a curriculum to enrich our Robotics Engineering, specifically, our course ECE8: Robot Automation: Intelligence, through Feedback Control through the use of STMicroelectronics drone kits. To this end, we developed a series of six ECE8 labs, which are detailed below, that to fit teaching modality during the pandemic, can be carried out remotely. The labs developed incoroporating STMicroelectronics drone kit are available at the following website: https://hybrid.soe.ucsc.edu/incorporating-stdrone-in-robotics-education

AFOSR DURIP: A Test Bed for Verification and Validation of Autonomous Systems with Hybrid Dynamics under Uncertainty

Principal Investigator(s): 
Prof. R. Sanfelice (UCSC), Prof. D. Venturi (UCSC), Prof. M. Wehner (U. Wisconsin)
Executive Summary: 

In conventional design processes, the design of the plant, the con- troller, the prototype as well as the certification of validity are products of consecutive phases of development utilizing distinct simulation, fabrication, and synthesis tools. Each phase produces an “optimal” solution, which is typically not jointly optimal for all phases. Phases need to be restarted from scratch if a new edge case or catastrophic failure is discovered at a later stage, and the interaction between failures, fabrication tolerances, and model inaccuracy is murky at best. To overcome these issues, and building from our effort as part of a current AFOSR grant, we propose to build a test bed implementing a coherent and provably rational methodology for automated design optimization, modeling, test and evaluation, and validation of autonomous systems with guaranteed levels of risk and robustness to uncertainty. The test bed will permit prototyping, experimentation, and redesign of autonomous systems and their components in an unified system. No such test bed is known yet to exist.

Pages