Simulation has been a necessary, safe testbed for robotics systems (RS). However, testing in simulation alone is not enough for robotic systems operating in close proximity, or interacting directly with, humans, because simulated humans are very limited. Furthermore, testing with real humans can be unsafe and costly. As recent advances in machine learning are being brought to physical robotic systems, how to collect data as well as evaluate them with human interactions safely yet realistically is a critical question. This paper presents a Mixed-Reality (MR) system toward human-centered development of robotic systems emphasizing benefits as a data collection and testbed tool. MR testbeds allow humans to interact with various levels of virtuality to maintain both realism and safety. We detail the advantages and limitations of these different levels of realism or virtualization, and report our MR-based RS testbed implemented using off-the-shelf MR devices with the Unity game engine and ROS. We demonstrate our testbed in a multi-robot, multi-person tracking and monitoring application. We share our vision and insights earned during the development and data collection.

Previous research has established redirected walking as a potential answer to exploring large virtual environments via natural locomotion within a limited physical space. However, much of the previous work has either focused on investigating human perception of redirected walking illusions or developing novel redirection techniques. In this paper, we take a broader look at the problem and formalize the concept of a complete redirected walking system. This work establishes the theoretical foundations for combining multiple redirection strategies into a unified framework known as adaptive redirection. This meta-strategy adapts based on the context, switching between a suite of strategies with a priori knowledge of their performance under the various circumstances. This paper also introduces a novel static planning strategy that optimizes gain parameters for a predetermined virtual path, known as the Combinatorially Optimized Pre-Planned Exploration Redirector (COPPER). We conducted a simulation-based experiment that demonstrates how adaptation rules can be determined empirically using machine learning, which involves partitioning the spectrum of contexts into regions according to the redirection strategy that performs best. Adaptive redirection provides a foundation for making redirected walking work in practice and can be extended to improve performance in the future as new techniques are integrated into the framework.

Developing effective strategies for redirected walking requires extensive evaluations across a variety of factors that influence performance. Because these large-scale experiments are often not practical with user studies, researchers have instead utilized simulations to systematically test different algorithm parameters, physical space configurations, and virtual walking paths. Although simulation offers an efficient way to evaluate redirected walking algorithms, it remains an open question whether this evaluation methodology is ecologically valid. In this paper, we investigate the interaction between locomotion behavior and redirection gains at a micro-level (across small path segments) and macro-level (across an entire experience). This examination involves analyzing data from real users and comparing algorithm performance metrics with a simulated user model. The results identify specific properties of user locomotion behavior that influence the application of redirected walking gains and resets. Overall, we found that the simulation provided a conservative estimate of the average performance with real users and observed that performance trends when comparing two redirected walking algorithms were preserved. In general, these results indicate that simulation is an empirically valid evaluation methodology for redirected walking algorithms.

Redirected walking techniques enable natural locomotion through immersive virtual environments that are considerably larger than the available real world walking space. However, the most effective strategy for steering the user remains an open question, as most previously presented algorithms simply redirect toward the center of the physical space. In this work, we present a theoretical framework that plans a walking path through a virtual environment and calculates the parameters for combining translation, rotation, and curvature gains such that the user can traverse a series of defined waypoints efficiently based on a utility function. This function minimizes the number of overt reorientations to avoid introducing potential breaks in presence. A notable advantage of this approach is that it leverages knowledge of the layout of both the physical and virtual environments to enhance the steering strategy.