A major challenge to deploying robots widely is navigation in human-populated environments, commonly referred to as social robot navigation . While the field of social navigation has advanced tremendously in recent years, the fair evaluation of algorithms that tackle social navigation remains hard because it involves not just robotic agents moving in static environments but also dynamic human agents and their perceptions of the appropriateness of robot behavior. In contrast, clear, repeatable, and accessible benchmarks have accelerated progress in fields like computer vision, natural language processing and traditional robot navigation by enabling researchers to fairly compare algorithms, revealing limitations of existing solutions and illuminating promising new directions. We believe the same approach can benefit social navigation. In this article, we pave the road toward common, widely accessible, and repeatable benchmarking criteria to evaluate social robot navigation. Our contributions include (a) a definition of a socially navigating robot as one that respects the principles of safety, comfort, legibility, politeness, social competency, agent understanding, proactivity, and responsiveness to context, (b) guidelines for the use of metrics, development of scenarios, benchmarks, datasets, and simulators to evaluate social navigation, and (c) a design of a social navigation metrics framework to make it easier to compare results from different simulators, robots, and datasets.

Understanding human perceptions of robot performance is crucial for designing socially intelligent robots that can adapt to human expectations. Current approaches often rely on surveys, which can disrupt ongoing human–robot interactions. As an alternative, we explore predicting people’s perceptions of robot performance using non-verbal behavioral cues and machine learning techniques. We contribute the SEAN TOGETHER Dataset consisting of observations of an interaction between a person and a mobile robot in Virtual Reality, together with perceptions of robot performance provided by users on a 5-point scale. We then analyze how well humans and supervised learning techniques can predict perceived robot performance based on different observation types (like facial expression and spatial behavior features). Our results suggest that facial expressions alone provide useful information, but in the navigation scenarios that we considered, reasoning about spatial features in context is critical for the prediction task. Also, supervised learning techniques outperformed humans’ predictions in most cases. Further, when predicting robot performance as a binary classification task on unseen users’ data, the \(F_{1}\) -Score of machine learning models more than doubled that of predictions on a 5-point scale. This suggested good generalization capabilities, particularly in identifying performance directionality over exact ratings. Based on these findings, we conducted a real-world demonstration where a mobile robot uses a machine learning model to predict how a human who follows it perceives it. Finally, we discuss the implications of our results for implementing these supervised learning models in real-world navigation. Our work paves the path to automatically enhancing robot behavior based on observations of users and inferences about their perceptions of a robot.

Long-range indoor navigation requires guiding robots with noisy sensors and controls through cluttered environments along paths that span a variety of buildings. We achieve this with PRM-RL, a hierarchical robot navigation method in which reinforcement learning (RL) agents that map noisy sensors to robot controls learn to solve short-range obstacle avoidance tasks, and then sampling-based planners map where these agents can reliably navigate in simulation; these roadmaps and agents are then deployed on robots, guiding them along the shortest path where the agents are likely to succeed. In this article, we use probabilistic roadmaps (PRMs) as the sampling-based planner, and AutoRL as the RL method in the indoor navigation context. We evaluate the method with a simulation for kinematic differential drive and kinodynamic car-like robots in several environments, and on differential-drive robots at three physical sites. Our results show that PRM-RL with AutoRL is more successful than several baselines, is robust to noise, and can guide robots over hundreds of meters in the face of noise and obstacles in both simulation and on robots, including over 5.8 km of physical robot navigation.

Deep Reinforcement Learning (RL) has recently emerged as a solution for moving obstacle avoidance. Deep RL learns to simultaneously predict obstacle motions and corresponding avoidance actions directly from robot sensors, even for obstacles with different dynamics models. However, deep RL methods typically cannot guarantee policy convergences, i.e., cannot provide probabilistic collision avoidance guarantees. In contrast, stochastic reachability (SR), a computationally expensive formal method that employs a known obstacle dynamics model, identifies the optimal avoidance policy and provides strict convergence guarantees. The availability of the optimal solution for versions of the moving obstacle problem provides a baseline to compare trained deep RL policies. In this paper, we compare the expected cumulative reward and actions of these policies to SR, and find the following. 1) The state-value function approximates the optimal collision probability well, thus explaining the high empirical performance. 2) RL policies deviate from the optimal significantly thus negatively impacting collision avoidance in some cases. 3) Evidence suggests that the deviation is caused, at least partially, by the actor net failing to approximate the action corresponding to the highest state-action value.

Identifying collision-free paths over long time windows in environments with stochastically moving obstacles is difficult, in part because long-term predictions of obstacle positions typically have low fidelity, and the region of possible obstacle occupancy is typically large. As a result, planning methods that are restricted to identifying paths with a low probability of collision may not be able to find a valid path. However, allowing paths with a higher probability of collision may limit detection of imminent collisions. In this paper, we present Dynamic Risk Tolerance (DRT), a framework that dynamically evaluates risk tolerance, a function which is formulated as a time-varying upper bound on the acceptable likelihood of collision for a given path. DRT is implemented with forward stochastic reachable sets to predict the exact distribution of obstacles in a scalable manner over an arbitrarily long time window. In effect, DRT identifies actions that balance risks posed by both near and far obstacles. We empirically compare DRT to other state of the art methods that are capable of generating real-time solutions in highly crowded environments, and demonstrate the success rates for DRT that is 46% higher than the best performing comparison method, in the most difficult problem tested.

This study presents Busy Beeway, a mobile game platform to investigate human-automation collaboration in dynamic environments. In Busy Beeway, users collaborate with automation to evade stochastically moving obstacles and reach a series of goals, in game levels of increasing difficulty. We are motivated by the need for reliable navigation aids in stochastic, dynamic environments, which are highly relevant for self-driving vehicles, UAVs, underwater and surface vehicles, and other applications. The proposed mobile game platform is agnostic to the particular algorithm underlying the autonomous system, can be used to evaluate both fully autonomous as well as human-in-the-loop systems, and is easily deployable, for large, remote user studies. This last element is key for rigorous study of human factors in navigation aids. Through a small 32--user study, we evaluate preliminary findings regarding the relative efficacy of collaborative and fully autonomous navigation, the relationship between success rate and users' learned trust in the automation (gathered via pre- and post-experiment surveys), and tolerance to error (for decisions made by the automation and by the user). This study validates the feasibility of Busy Beeway as a platform for human subject studies on human-automation collaboration, and suggests directions for future research in human-aided planning in difficult environments.

One of the primary challenges for autonomous robotics in uncertain and dynamic environments is planning and executing a collision-free path. Hybrid dynamic obstacles present an even greater challenge as the obstacles can change dynamics without warning and potentially invalidate paths. Artificial potential field (APF)-based techniques have shown great promise in successful path planning in highly dynamic environments due to their low cost at runtime. We utilize the APF framework for runtime planning but leverage a formal validation method, Stochastic Reachable (SR) sets, to generate accurate potential fields for moving obstacles. A small number of SR sets are computed a priori, then used to generate a potential field that represents the obstacle's stochastic motion for online path planning. Our method is novel and scales well with the number of obstacles, maintaining a relatively high probability of reaching the goal without collision, as compared to other traditional Gaussian APF methods. Here, we demonstrate our method with up to 900 hybrid dynamic obstacles and show that it outperforms the traditional Gaussian APF method by up to 60% in the holonomic case and up to 20% in the unicycle case.

Manual derivation of optimal robot motions for task completion is difficult, especially when a robot is required to balance its actions between opposing preferences. One solution has been proposed to automatically learn near optimal motions with Reinforcement Learning (RL). This has been successful for several tasks including swing-free UAV flight, table tennis, and autonomous driving. However, high-dimensional problems remain a challenge. We address this dimensionality constraint with PrEference Appraisal Reinforcement Learning (PEARL), which solves tasks with opposing preferences for acceleration controlled robots. PEARL projects the high-dimensional continuous robot state space to a low dimensional preference feature space resulting in efficient and adaptable planning. We demonstrate that on a dynamic obstacle avoidance robotic task, a single learning on a much simpler problem performs real-time decision-making for significantly larger, high-dimensional problems working in unbounded continuous states and actions. We trained the agent with 4 static obstacles, while the trained agent avoids up to 900 moving obstacles with complex hybrid stochastic obstacle dynamics in a highly constrained space using only limited information about the environment. We compare these tasks to traditional, often manually tuned solutions for these high-dimensional problems.

Motion planning in stochastic dynamic uncertain environments is critical in several applications such as human interacting robots, autonomous vehicles and assistive robots. In order to address these complex applications, several methods have been developed. The most successful methods often predict future obstacle locations in order identify collision-free paths. Since prediction can be computationally expensive, offline computations are commonly used, and simplifications such as the inability to consider the dynamics of interacting obstacles or possible stochastic dynamics are often applied. Online methods can be preferable to simulate potential obstacle interactions, but recent methods have been restricted to Gaussian interaction processes and uncertainty. In this paper we present an online motion planning method, Runtime Stochastic Ensemble Simulation (Runtime SES) planning, an inexpensive method for predicting obstacle motion with generic stochastic dynamics while maintaining a high planning success rate despite the potential presence of obstacle position error. Runtime SES planning evaluates the likelihood of collision for any state-time coordinate around the robot by performing Monte Carlo simulations online. This prediction is used to construct a customized Rapidly Exploring Random Tree (RRT) in order to quickly identify paths that avoid obstacles while moving toward a goal. We demonstrate Runtime SES planning in problems that benefit from online predictions, environments with strongly-interacting obstacles with stochastic dynamics and positional error. Through experiments that explore the impact of various parametrizations, robot dynamics and obstacle interaction models, we show that real-time capable planning with a high success rate is achievable in several complex environments.

Highly dynamic environments pose a particular challenge for motion planning due to the need for constant evaluation or validation of plans. However, due to the wide range of applications, an algorithm to safely plan in the presence of moving obstacles is required. In this paper, we propose a novel technique that provides computationally efficient planning solutions in environments with static obstacles and several dynamic obstacles with stochastic motions. Path-Guided APF-SR works by first applying a sampling-based technique to identify a valid, collision-free path in the presence of static obstacles. Then, an artificial potential field planning method is used to safely navigate through the moving obstacles using the path as an attractive intermediate goal bias. In order to improve the safety of the artificial potential field, repulsive potential fields around moving obstacles are calculated with stochastic reachable sets, a method previously shown to significantly improve planning success in highly dynamic environments. We show that Path-Guided APF-SR outperforms other methods that have high planning success in environments with 300 stochastically moving obstacles. Furthermore, planning is achievable in environments in which previously developed methods have failed.

Motion planning in stochastic dynamic environments is difficult due to the need for constant plan adjustment caused by the uncertainty of the environment. There are many motion planning problems, including flight coordination and autonomous vehicles, that require an algorithm to predict obstacle motion and plan safely. In this paper, we propose Stochastic Ensemble Simulation (SES)-based planning, a novel framework to efficiently predict and produce safe trajectories in the presence of stochastic obstacles. The stochastic obstacles can be introduced in several ways including stochastic motion or position/speed uncertainty. SES-based planning works by first predicting an obstacle's future position offline through an ensemble of Monte Carlo simulations. These runs simulate the stochastic obstacle dynamics and store the simulation results in temporal snapshots of predicted positions. An online planner then uses this output to identify a predicted collision-free direct path to the goal. If the direct path is not expected to be collision-free, a more expensive tree-based planner is used. Our experiments show SES-based planning outperforms other methods that have high planning success in environments with 900 stochastically moving obstacles. Furthermore, our method plans trajectories with an 80% success rate for a 7 DOF robot in an environment with 250 stochastic moving obstacles and 50 obstacles with speed/position uncertainty. This complex problem is currently beyond the capability of several comparison methods.