Robotic assistive devices offer unique benefits in industrial and medical settings augmenting purely human manipulation, such as precise micro-movements, and well-defined and concentrated force application with example use cases including soft-tissue inspection or surface polishing. Despite the abundance of literature focusing on laboratory demonstrations and experiments, the adoption of such technologies in industrial and medical settings remains limited due to concerns over safety, reliability, and training requirements. This paper introduces a reliable and adaptable control system framework for teleoperation devices with haptic force-feedback input, utilizing the Optimization-based Task Specification (OpTaS) Library. Our proposed framework addresses the aforementioned concerns by offering a control system with minimal training requirements and easily customizable safety constraints, aimed at enhancing deployability. A key contribution of the proposed control system is the introduction of inverse differential kinematics to real-time teleoperation, which enables the definition of safety constraints on the output of the controller providing an inherently safer implementation in contrast to fully inverse kinematic solutions.

A Control System Framework for Robust Deployability of Teleoperation Devices in Shared Workspaces

Neural-Fly is a learning-based control method that allows rapid online learning by incorporating a pre-trained representations of unmodeled aerodynamics with adaptive control.
Not is Neural-Fly robust to different tasks, environments, and drones, but also it builds off a standard control architecture, creating a straightforward way to begin integrating learning-based control into safety critical applications.
In this workshop, we will present the latest results for Neural-Fly on field robots, discuss implementation considerations, and show case other applications of Neural-Fly, such as fault tolerant control. 

Deploying Neural-Fly in the Field

Sidewalks are increasingly being shared by autonomous mobile robots, creating a safety-sensitive environment that lacks explicit safety-focused standards. Establishing such standards is an essential prerequisite for progressing these vehicles from the laboratory to lager-scale, widespread use. This ongoing research presented here aims to fill this gap by exploring standards in related areas, assessing their applicability, and examining possible derived specifications adapted to mobile robots operating on sidewalks and similar traffic environments. As part of our methodology, we draw upon findings from the development of two delivery robot prototype vehicles and a literature review. By establishing a foundation for the design and assessment of future standard drafts, our research supports the broader adoption of mobile robots and contributes to safer urban sidewalks.

Bridging the Normative Gap: Standardization for Sidewalk Robots in a World of Self-Driving Cars, Personal Robots and Automated Industrial Vehicles

This extended abstract underscores the prospect and fundamental challenges of deploying robotic radiation detectors in nuclear safeguards and arms control applications, where safety and security concerns are of utmost importance. We propose a performant, minimal-knowledge approach which addresses information security constraints, with specific emphasis on the challenge of confirming the absence of undeclared radioactive sources. We demonstrate a random walk process which requires no a priori knowledge of the environment and certifies the absence or presence of sources without revealing any a posteriori information. We improve the convergence of this method by incorporating directional radiation measurements. The finite time and physical safety limitations may be addressed by limited, rather than minimal, knowledge approaches. The methods discussed illuminate a spectrum of limited-information contexts which have not yet been thoroughly explored in this application space.

Realizable Deployment of Limited-Knowledge Robotic Inspectors for Nuclear Verification

Despite the widespread use of Autonomous Unmanned Aerial Vehicles (UAVs) in various outdoor applications, their deployment in indoor and GPS-denied environments—such as under dense forest canopies—continues to pose significant challenges. Robust trajectory planning, accurate state estimation, and effective obstacle avoidance are crucial to successful mission completion in these complex, dynamic, and unstructured settings. In this work, we present our open source autonomy stack for aerial robots operating in GPS-denied environments. We specifically focus on two critical components: (1) compensation for inaccurate state estimation and (2) trajectory planning in uncertain scenarios. Using these techniques, our systems are capable of executing long-range missions in challenging environments. We hope that the tools and methodologies introduced in this study will accelerate the adoption and safe deployment of UAVs in GPS-denied conditions.

Open Source Tools for Deployment of GPS-Denied Autonomous UAVs in Real-World Applications

Building reliable Cyber-Physical Systems (CPS) often requires formal verification, which is capable of producing the rigorous safety guarantees necessary to provide confidence in its design. However, verification may be challenging to perform on real-world systems, and as a result, a simplified model may be verified instead. This results in a disconnect between verification and reality, leading to diminished confidence in the verification result. We have developed a method that enables CPS designers to both formally verify and implement CPS software in a single language. By combining the industry-tested modeling and compilation of synchronous programming with the verification rigor of refinement types, we strive to give CPS designers additional confidence in designing reliable, real-world CPS. 

Bridging the Cyber and Physical with a Verifiable, Executable Language for Robotics

This study proposes an automated virtual test creation pipeline that replicates real-world environments in simulation for autonomous vehicle development. By integrating publicly available tools, the pipeline accelerates AV testing with minimal human intervention. Our approach develops real-world maps for open-source simulators like CARLA and LGSVL, facilitating testing and development of most exist- ing autonomous driving controllers, including Autoware and Apollo. This study outlines the methodology and results of our approach, highlighting its potential to significantly enhance the scalability and efficiency of autonomous vehicle testing and development.