Accurately predicting human behaviors is crucial for mobile robots operating in human-populated environments. While prior research primarily focuses on predicting actions in single-human scenarios from an egocentric view, several robotic applications require understanding multiple human behaviors from a third-person perspective. To this end, we present CAMP-VLM (Context-Aware Multi-human behavior Prediction): a Vision Language Model (VLM)-based framework that incorporates contextual features from visual input and spatial awareness from scene graphs to enhance prediction of humans-scene interactions. Due to the lack of suitable datasets for multi-human behavior prediction from an observer view, we perform fine-tuning of CAMP-VLM with synthetic human behavior data generated by a photorealistic simulator, and evaluate the resulting models on both synthetic and real-world sequences to assess their generalization capabilities. Leveraging Supervised Fine-Tuning (SFT) and Direct Preference Optimization (DPO), CAMP-VLM outperforms the best-performing baseline by up to 66.9% in prediction accuracy.

Accurate human activity and trajectory prediction are crucial for ensuring safe and reliable human-robot interactions in dynamic environments, such as industrial settings, with mobile robots. Datasets with fine-grained action labels for moving people in industrial environments with mobile robots are scarce, as most existing datasets focus on social navigation in public spaces. This paper introduces the THÖR-MAGNI Act dataset, a substantial extension of the THÖR-MAGNI dataset, which captures participant movements alongside robots in diverse semantic and spatial contexts. THOR-MAGNI Act provides 8.3 hours of manually labeled participant actions derived from egocentric videos recorded via eye-tracking glasses. These actions, aligned with the provided THOR-MAGNI motion cues, follow a long-tailed distribution with diversified acceleration, velocity, and navigation distance profiles. We demonstrate the utility of THÖR-MAGNI Act for two tasks: action-conditioned trajectory prediction and joint action and trajectory prediction. We propose two efficient transformer-based models that outperform the baselines to address these tasks. These results underscore the potential of THOR-MAGNI Act to develop predictive models for enhanced human-robot interaction in complex environments.

Recent advancements in Large Language Models (LLMs) have sparked a revolution across many research fields. In robotics, the integration of common-sense knowledge from LLMs into task and motion planning has drastically advanced the field by unlocking unprecedented levels of context awareness. Despite their vast collection of knowledge, large language models may generate infeasible plans due to hal-lucinations or missing domain information. To address these challenges and improve plan feasibility and computational efficiency, we introduce DELTA, a novel LLM-informed task planning approach. By using scene graphs as environment representations within LLMs, DELTA achieves rapid generation of precise planning problem descriptions. To enhance planning performance, DELTA decomposes long-term task goals with LLMs into an autoregressive sequence of sub-goals, enabling automated task planners to efficiently solve complex problems. In our extensive evaluation, we show that DELTA enables an efficient and fully automatic task planning pipeline, achieving higher planning success rates and significantly shorter planning times compared to the state of the art. Project webpage: https://delta-llm.github.io/

Maps of dynamics are effective representations of motion patterns learned from prior observations, with recent research demonstrating their ability to enhance various downstream tasks such as human-aware robot navigation, long-term human motion prediction, and robot localization. Current advancements have primarily concentrated on methods for learning maps of human flow in environments where the flow is static, i.e., not assumed to change over time. In this paper we propose an online update method of the CLiFF-map (an advanced map of dynamics type that models motion patterns as velocity and orientation mixtures) to actively detect and adapt to human flow changes. As new observations are collected, our goal is to update a CLiFF-map to effectively and accurately integrate them, while retaining relevant historic motion patterns. The proposed online update method maintains a probabilistic representation in each observed location, updating parameters by continuously tracking sufficient statistics. In experiments using both synthetic and real-world datasets, we show that our method is able to maintain accurate representations of human motion dynamics, contributing to high performance flow-compliant planning downstream tasks, while being orders of magnitude faster than the comparable baselines.

Rare, yet critical, scenarios pose a significant challenge in testing and evaluating autonomous driving planners. Relying solely on real-world driving scenes requires collecting massive datasets to capture these scenarios. While automatic generation of traffic scenarios appears promising, data-driven models require extensive training data and often lack fine-grained control over the output. Moreover, generating novel scenarios from scratch can introduce a distributional shift from the original training scenes which undermines the validity of evaluations especially for learning-based planners. To sidestep this, recent work proposes to generate challenging scenarios by augmenting original scenarios from the test set. However, this involves the manual augmentation of scenarios by domain experts. An approach that is unable to meet the demands for scale in the evaluation of self-driving systems. Therefore, this paper introduces a novel LLM-agent based framework for augmenting real-world traffic scenarios using natural language descriptions, addressing the limitations of existing methods. A key innovation is the use of an agentic design, enabling fine-grained control over the output and maintaining high performance even with smaller, cost-effective LLMs. Extensive human expert evaluation demonstrates our framework’s ability to accurately adhere to user intent, generating high quality augmented scenarios comparable to those created manually.

Predicting human behavior in shared environments is crucial for safe and efficient human-robot interaction. Traditional data-driven methods to that end are pre-trained on domain-specific datasets, activity types, and prediction horizons. In contrast, the recent breakthroughs in Large Language Models (LLMs) promise open-ended cross-domain generalization to describe various human activities and make predictions in any context. In particular, Multimodal LLMs (MLLMs) are able to integrate information from various sources, achieving more contextual awareness and improved scene understanding. The difficulty in applying general-purpose MLLMs directly for prediction stems from their limited capacity for processing large input sequences, sensitivity to prompt design, and expensive fine-tuning. In this paper, we present a systematic analysis of applying pre-trained MLLMs for context-aware human behavior prediction. To this end, we introduce a modular multimodal human activity prediction framework that allows us to benchmark various MLLMs, input variations, In-Context Learning (ICL), and autoregressive techniques. Our evaluation indicates that the best-performing framework configuration is able to reach 92.8% semantic similarity and 66.1% exact label accuracy in predicting human behaviors in the target frame. Project webpage: https://cap-mllm.github.io/

In the active landscape of AI research, decentralised learning is gaining momentum. Decentralised learning allows individual nodes to keep data locally where they are generated and to share knowledge extracted from local data among themselves through an interactive process of collaborative refinement. This paradigm supports scenarios where data cannot leave the data owner node due to privacy or sovereignty reasons or real-time constraints imposing proximity of models to locations where inference has to be carried out. The distributed nature of decentralised learning implies significant new research challenges with respect to centralised learning. Among them, in this paper, we focus on robustness issues. Specifically, we study the effect of nodes’ disruption on the collective learning process. Assuming a given percentage of “central” nodes disappear from the network, we focus on different cases, characterised by (i) different distributions of data across nodes and (ii) different times when disruption occurs with respect to the start of the collaborative learning task. Through these configurations, we are able to show the non-trivial interplay between the properties of the network connecting nodes, the persistence of knowledge acquired collectively before disruption or lack thereof, and the effect of data availability pre- and post-disruption. Our results show that decentralised learning processes are remarkably robust to network disruption. As long as even minimum amounts of data remain available somewhere in the network, the learning process is able to recover from disruptions and achieve significant classification accuracy. This clearly varies depending on the remaining connectivity after disruption, but we show that even nodes that remain completely isolated can retain significant knowledge acquired before the disruption.

Coordinating a fleet of robots in unstructured, human-shared environments is challenging. Human behavior is hard to predict, and its uncertainty impacts the performance of the robotic fleet. Various multi-robot planning and coordination algorithms have been proposed, including Multi-Agent Path Finding (MAPF) methods to precedence-based algorithms. However, it is still unclear how human presence impacts different coordination strategies in both simulated environments and the real world. With the goal of studying and further improving multi-robot planning capabilities in those settings, we propose a method to develop and benchmark different multi-robot coordination algorithms in realistic, unstructured and human-shared environments. To this end, we introduce a multi-robot benchmark framework that is based on state-of-the-art open-source navigation and simulation frameworks and can use different types of robots, environments and human motion models. We show a possible application of the benchmark framework with two different environments and three centralized coordination methods (two MAPF algorithms and a loosely-coupled coordination method based on precedence constraints). We evaluate each environment for different human densities to investigate its impact on each coordination method. We also present preliminary results that show how informing each coordination method about human presence can help the coordination method to find faster paths for the robots.

Visual Reinforcement Learning (RL) methods often require extensive amounts of data. As opposed to model-free RL, model-based RL (MBRL) offers a potential solution with efficient data utilization through planning. Additionally, RL lacks generalization capabilities for real-world tasks. Prior work has shown that incorporating pre-trained visual representations (PVRs) enhances sample efficiency and generalization. While PVRs have been extensively studied in the context of model-free RL, their potential in MBRL remains largely unexplored. In this paper, we benchmark a set of PVRs on challenging control tasks in a model-based RL setting. We investigate the data efficiency, generalization capabilities, and the impact of different properties of PVRs on the performance of model-based agents. Our results, perhaps surprisingly, reveal that for MBRL current PVRs are not more sample efficient than learning representations from scratch, and that they do not generalize better to out-of-distribution (OOD) settings. To explain this, we analyze the quality of the trained dynamics model. Furthermore, we show that data diversity and network architecture are the most important contributors to OOD generalization performance.

Fully decentralized learning is gaining momentum for training AI models at the Internet’s edge, addressing infrastructure challenges and privacy concerns. In a decentralized machine learning system, data is distributed across multiple nodes, with each node training a local model based on its respective dataset. The local models are then shared and combined to form a global model capable of making accurate predictions on new data. Our exploration focuses on how different types of network structures influence the spreading of knowledge - the process by which nodes incorporate insights gained from learning patterns in data available on other nodes across the network. Specifically, this study investigates the intricate interplay between network structure and learning performance using three network topologies and six data distribution methods. These methods consider different vertex properties, including degree centrality, betweenness centrality, and clustering coefficient, along with whether nodes exhibit high or low values of these metrics. Our findings underscore the significance of global centrality metrics (degree, betweenness) in correlating with learning performance, while local clustering proves less predictive. We highlight the challenges in transferring knowledge from peripheral to central nodes, attributed to a dilution effect during model aggregation. Additionally, we observe that central nodes exert a pull effect, facilitating the spread of knowledge. In examining degree distribution, hubs in Barabási–Albert networks positively impact learning for central nodes but exacerbate dilution when knowledge originates from peripheral nodes. Finally, we demonstrate the formidable challenge of knowledge circulation outside of segregated communities, and discuss the impact of class cross-correlations.

The trajectory planning for a fleet of automated guided vehicles (AGVs) on a roadmap is commonly referred to as the multi-agent path finding (MAPF) problem, the solution to which dictates each AGV's spatial and temporal location until it reaches its goal without collision. When executing MAPF plans in dynamic workspaces, AGVs can be frequently delayed, e.g., due to encounters with humans or third-party vehicles. If the remainder of the AGVs keeps following their individual plans, synchrony of the fleet is lost and some AGVs may pass through roadmap intersections in a different order than originally planned. Although this could reduce the cumulative route completion time of the AGVs, generally, a change in the original ordering can cause conflicts, such as deadlocks. In practice, synchrony is therefore often enforced by using a MAPF execution policy employing, e.g., an action dependency graph (ADG) to maintain ordering. To safely re-order without introducing deadlocks, we present the concept of the switchable action dependency graph (SADG). Using the SADG, we formulate a comparatively low-dimensional mixed-integer linear program that repeatedly re-orders AGVs in a recursively feasible manner, thus maintaining deadlock-free guarantees, while dynamically minimizing the cumulative route completion time of all AGVs. Various simulations validate the efficiency of our approach when compared to the original ADG method as well as robust MAPF solution approaches.

Several successful approaches exist for solving the complex problem of multi-robot planning and coordination. Due to the lack of adequate benchmarking tools, comparing these approaches and judging their suitability for use in realistic scenarios is currently difficult. Therefore, we propose an open-source benchmark suite that aims to close this gap. Unlike existing benchmarks, our approach uses full-stack multi-robot navigation systems in realistic 3D simulated environments from the intralogistic and household domains. Using the open-source frameworks ROS 2, Gazebo and RMF allows the user to add other robot platforms easily. The framework provides easy-to-use abstractions, typical metrics and interfaces to several established planning libraries for multi-robot systems. With all these features, our framework successfully aids practitioners and researchers in comparing multi-robot planning and coordination systems to the state of the art. Our experiments show how the proposed benchmark simplifies gaining insights on relevant close to real-life robotics use cases.

We investigate the task of object goal navigation in unknown environments where a target object is given as a semantic label (e.g. find a couch). This task is challenging as it requires the robot to consider the semantic context in diverse settings (e.g. TVs are often nearby couches). Most of the prior work tackles this problem under the assumption of a discrete action policy whereas we present an approach with continuous control which brings it closer to real world applications. In this paper, we use information-theoretic model predictive control on dense cost maps to bring object goal navigation closer to real robots with kinodynamic constraints. We propose a deep neural network framework to learn cost maps that encode semantic context and guide the robot towards the target object. We also present a novel way of fusing mid-level visual representations in our architecture to provide additional semantic cues for cost map prediction. The experiments show that our method leads to more efficient and accurate goal navigation with higher quality paths than the reported baselines. The results also indicate the importance of mid-level representations for navigation by improving the success rate by 8 percentage points.

For robots navigating in dynamic environments, exploiting and understanding uncertain human motion prediction is key to generate efficient, safe and legible actions. The robot may perform poorly and cause hindrances if it does not reason over possible, multi-modal future social interactions. With the goal of enhancing autonomous navigation in cluttered environments, we propose a novel formulation for nonlinear model predictive control including multi-modal predictions of human motion. As a result, our approach leads to less conservative, smooth and intuitive human-aware navigation with reduced risk of collisions, and shows a good balance between task efficiency, collision avoidance and human comfort. To show its effectiveness, we compare our approach against the state of the art in crowded simulated environments, and with real-world human motion data from the THOR dataset. This comparison shows that we are able to improve task efficiency, keep a larger distance to humans and significantly reduce the collision time, when navigating in cluttered dynamic environ-ments. Furthermore, the method is shown to work robustly with different state-of-the-art human motion predictors.

Human motion prediction is important for mobile service robots and intelligent vehicles to operate safely and smoothly around people. The more accurate predictions are, particularly over extended periods of time, the better a system can, e.g., assess collision risks and plan ahead. In this paper, we propose to exploit maps of dynamics (MoDs, a class of general representations of place-dependent spatial motion patterns, learned from prior observations) for long-term human motion prediction (LHMP). We present a new MoD-informed human motion prediction approach, named CLiFF-LHMP, which is data efficient, explainable, and insensitive to errors from an upstream tracking system. Our approach uses CLiFF -map, a specific MoD trained with human motion data recorded in the same environment. We bias a constant velocity prediction with samples from the CLiFF-map to generate multi-modal trajectory predictions. In two public datasets we show that this algorithm outperforms the state of the art for predictions over very extended periods of time, achieving 45 % more accurate prediction performance at 50s compared to the baseline.

Human motion trajectory prediction, an essential task for autonomous systems in many domains, has been on the rise in recent years. With a multitude of new methods proposed by different communities, the lack of standardized benchmarks and objective comparisons is increasingly becoming a major limitation to assess progress and guide further research. Existing benchmarks are limited in their scope and flexibility to conduct relevant experiments and to account for contextual cues of agents and environments. In this paper we present Atlas, a benchmark to systematically evaluate human motion trajectory prediction algorithms in a unified framework. Atlas offers data preprocessing functions, hyperparameter optimization, comes with popular datasets and has the flexibility to setup and conduct underexplored yet relevant experiments to analyze a method’s accuracy and robustness. In an example application of Atlas, we compare five popular model- and learning-based predictors and find that, when properly applied, early physics-based approaches are still remarkably competitive. Such results confirm the necessity of benchmarks like Atlas.

Even though mobile robots have been around for decades, trajectory optimization and continuous time collision avoidance remain subject of active research. Existing methods trade off between path quality, computational complexity, and kinodynamic feasibility. This work approaches the problem using a nonlinear model predictive control (NMPC) framework, that is based on a novel convex inner approximation of the collision avoidance constraint. The proposed Convex Inner ApprOximation (CIAO) method finds kinodynamically feasible and continuous time collision free trajectories, in few iterations, typically one. For a feasible initialization, the approach is guaranteed to find a feasible solution, i.e. it preserves feasibility. Our experimental evaluation shows that CIAO outperforms state of the art baselines in terms of planning efficiency and path quality. Experiments show that it also efficiently scales to high-dimensional systems. Furthermore real-world experiments demonstrate its capability of unifying trajectory optimization and tracking for safe motion planning in dynamic environments.

The problem of minimizing cost in nonlinear control systems with uncertainties or disturbances remains a major challenge. Model predictive control (MPC), and in particular sampling-based MPC has recently shown great success in complex domains such as aggressive driving with highly nonlinear dynamics. Sampling-based methods rely on a prior distribution to generate samples in the first place. Obviously, the choice of this distribution highly influences efficiency of the controller. Existing approaches such as sampling around the control trajectory of the previous time step perform suboptimally, especially in multi-modal or highly dynamic settings. In this work, we therefore propose to learn models that generate samples in low-cost areas of the state-space, conditioned on the environment and on contextual information of the task to solve. By using generative models as an informed sampling distribution, our approach exploits guidance from the learned models and at the same time maintains robustness properties of the MPC methods. We use Conditional Variational Autoencoders (CVAE) to learn distributions that imitate samples from a training dataset containing optimized controls. An extensive evaluation in the autonomous navigation domain suggests that replacing previous sampling schemes with our learned models considerably improves performance in terms of path quality and planning efficiency.

Planning smooth trajectories is important for the safe, efficient and comfortable operation of mobile robots, such as wheeled robots moving in crowded environments or cars moving at high speed. Asymptotically optimal sampling-based motion planners can be used to generate such trajectories. However, to achieve the necessary efficiency for the realtime operation of robots, one often uses their initial feasible trajectories or the trajectories of non-optimal motion planners instead, typically after a post-smoothing step. We propose a gradient-informed post-smoothing algorithm, called GRIPS, that deforms given trajectories by locally optimizing the placement of vertices while satisfying the system's kinodynamic constraints. We show experimentally that GRIPS typically produces trajectories of significantly smaller length and higher smoothness than several existing post-smoothing algorithms.

The ability to perceive and predict future positions of dynamic objects is essential for mobile robots and intelligent vehicles in dynamic environments. In this paper, we present a novel planning-based approach for long-term human motion prediction that accounts for local interactions and can accurately predict joint motion of multiple agents. Long-term predictions are handled using an MDP formulation that computes a set of stochastic motion policies. To obtain distributions over future motion trajectories, we sample the policies with a weighted random walk algorithm in which each person is locally influenced by social forces from other nearby agents. Unlike related work, the algorithm is environment-aware, can account for individual agent velocities, requires no training phase and makes joint predictions for multiple agents. Experiments in simulation and with real data show that our method makes more accurate predictions than two state-of-the-art methods in terms of probabilistic and geometrical performance measures.

Accurate long-term prediction of human motion in populated spaces is an important but difficult task for mobile robots and intelligent vehicles. What makes this task challenging is that human motion is influenced by a large variety of factors including the person's intention, the presence, attributes, actions, social relations and social norms of other surrounding agents, and the geometry and semantics of the environment. In this paper, we consider the problem of computing human motion predictions that account for such factors. We formulate the task as an MDP planning problem with stochastic policies and propose a weighted random walk algorithm in which each agent is locally influenced by social forces from other nearby agents. The novelty of this paper is that we incorporate social grouping information into the prediction process reflecting the soft formation constraints that groups typically impose to their members' motion. We show that our method makes more accurate predictions than three state-of-the-art methods in terms of probabilistic and geometrical performance metrics.

In this paper we address the problem of flow-aware trajectory planning in dynamic environments considering flow model uncertainty. Flow-aware planning aims to plan trajectories that adhere to existing flow motion patterns in the environment, with the goal to make robots more efficient, less intrusive and safer. We use a statistical model called CLiFF-map that can map flow patterns for both continuous media and discrete objects. We propose novel cost and biasing functions for an RRT* planning algorithm, which exploits all the information available in the CLiFF-map model, including uncertainties due to flow variability or partial observability. Qualitatively, a benefit of our approach is that it can also be tuned to yield trajectories with different qualities such as exploratory or cautious, depending on application requirements. Quantitatively, we demonstrate that our approach produces more flow-compliant trajectories, compared to two baselines.

We present a mobile robot motion planning approach under kinodynamic constraints that exploits learned perception priors in the form of continuous Gaussian mixture fields. Our Gaussian mixture fields are statistical multi-modal motion models of discrete objects or continuous media in the environment that encode e.g. the dynamics of air or pedestrian flows. We approach this task using a recently proposed circular linear flow field map based on semi-wrapped GMMs whose mixture components guide sampling and rewiring in an RRT* algorithm using a steer function for non-holonomic mobile robots. In our experiments with three alternative baselines, we show that this combination allows the planner to very efficiently generate high-quality solutions in terms of path smoothness, path length as well as natural yet minimum control effort motions through multi-modal representations of Gaussian mixture fields.

RRT and RRT* have become popular planning techniques, in particular for high-dimensional systems such as wheeled robots with complex nonholonomic constraints. Their planning times, however, can scale poorly for such robots, which has motivated researchers to study hierarchical techniques that grow the RRT trees in more focused ways. Along this line, we introduce Theta*-RRT that hierarchically combines (discrete) any-angle search with (continuous) RRT motion planning for nonholonomic wheeled robots. Theta*-RRT is a variant of RRT that generates a trajectory by expanding a tree of geodesics toward sampled states whose distribution summarizes geometric information of the any-angle path. We show experimentally, for both a differential drive system and a high-dimensional truck-and-trailer system, that Theta*-RRT finds shorter trajectories significantly faster than four baseline planners (RRT, A*-RRT, RRT*, A*-RRT*) without loss of smoothness, while A*-RRT* and RRT* (and thus also Informed RRT*) fail to generate a first trajectory sufficiently fast in environments with complex nonholonomic constraints. We also prove that Theta*-RRT retains the probabilistic completeness of RRT for all small-time controllable systems that use an analytical steer function.

The distance metric is a key component in RRT-based motion planning that deeply affects coverage of the state space, path quality and planning time. With the goal to speed up planning time, we introduce a learning approach to approximate the distance metric for RRT-based planners. By exploiting a novel steer function which solves the two-point boundary value problem for wheeled mobile robots, we train a simple nonlinear parametric model with constant-time inference that is shown to predict distances accurately in terms of regression and ranking performance. In an extensive analysis we compare our approach to an Euclidean distance baseline, consider four alternative regression models and study the impact of domain-specific feature expansion. The learning approach is shown to be faster in planning time by several factors at negligible loss of path quality.

In this paper we introduce a novel RRT extend function for wheeled mobile robots. The approach computes closed-loop forward simulations based on the kinematic model of the robot and enables the planner to efficiently generate smooth and feasible paths that connect any pairs of states. We extend the control law of an existing discontinuous state feedback controller to make it usable as an RRT extend function and prove that all relevant stability properties are retained. We study the properties of the new approach as extender for RRT and RRT* and compare it systematically to a spline-based approach and a large and small set of motion primitives. The results show that our approach generally produces smoother paths to the goal in less time with smaller trees. For RRT*, the approach produces also the shortest paths and achieves the lowest cost solutions when given more planning time.