Autonomous exploration is a fundamental problem for various applications of unmanned aerial vehicles (UAVs). Recently, LiDAR-based exploration has gained significant attention due to its ability to generate high-precision point cloud maps of large-scale environments. While the point clouds are inherently informative for navigation, many existing exploration methods still rely on additional, often expensive, environmental representations. This reliance stems from two main reasons: the need for frontier detection or information gain computation, which typically depends on memory-intensive occupancy grid maps, and the high computational complexity of path planning directly on point clouds, primarily due to costly collision checking. To address these limitations, we present EPIC, a lightweight LiDAR-based UAV exploration framework that directly exploits point cloud data to explore large-scale environments. EPIC introduces a novel observation map derived directly from the quality of point clouds, eliminating the need for global occupancy grid maps while preserving comprehensive exploration capabilities. We also propose an incremental topological graph construction method operating directly on point clouds, enabling real-time path planning in large-scale environments. Leveraging these components, we build a hierarchical planning framework that generates agile and energy-efficient trajectories, achieving significantly reduced memory consumption and computation time compared to most existing methods. Extensive simulations and real-world experiments demonstrate that EPIC achieves faster exploration while significantly reducing memory consumption compared to state-of-the-art methods.

This paper introduces a novel approach that integrates future closest point predictions into the distance constraints of a collision avoidance controller, leveraging convex hulls with closest point distance calculations. By addressing abrupt shifts in closest points, this method effectively reduces collision risks and enhances controller performance. Applied to an Image Guided Therapy robot and validated through simulations and user experiments, the framework demonstrates improved distance prediction accuracy, smoother trajectories, and safer navigation near obstacles.

Robot manipulation usually involves multiple different manipulation primitives, such as Push and Pivot, leading to hybrid and long-horizon characteristics. This poses significant challenges to most planning and control approaches. Instead of treating long-horizon manipulation as a whole, it can be decomposed into several simple manipulation primitives and then sequenced using PDDL planners [1, 2, 3] or Large Language Models [4, 5]. Although such manipulation primitives usually have lowto-medium-dimensional state and action spaces, the breaking and establishment of contact make it tough for most motion planning techniques. Gradient-based techniques suffer from vanishing gradients when contact breaks, while sampling-based techniques struggle with the combinatorial complexity of multiple contact modes, i.e., sticking and sliding. This leads to time-consuming online replanning in the real world for contact-rich manipulation, limiting the real-time reactiveness of robots in coping with uncertainties and disturbances. Learning manipulation primitives that can quickly react to the surroundings, therefore, makes a lot of sense. Since the learned manipulation primitives will be sequenced by symbolic planners, which have no information about the geometric/motion level, the learned manipulation primitive should be robust to diverse instances with varied physical parameters, such as shape, mass, and friction coefficient. For example, once the push primitive is scheduled by the high-level symbolic planner, it should be able to Figure 2: Illustration of DA, DR and DC.

Autonomous navigation of a mobile robot is a challenging task which requires ability of mapping, localization, path planning and path following. Conventional mapping methods build a dense metric map like an occupancy grid, which is affected by odometry error accumulation and consumes a lot of memory and computations in large environments. Another approach to mapping is the usage of topological properties, e.g. adjacency of locations in the environment. Topological maps are less prone to odometry error accumulation and high resources consumption, and also enable fast path planning because of the graph sparsity. Based on this idea, we proposed NavTopo - a full navigation pipeline based on topological map and two-level path planning. The pipeline localizes in the graph by matching neural network descriptors and 2D projections of the input point clouds, which significantly reduces memory consumption compared to metric and topological point cloud-based approaches. We test our approach in a large indoor photo-relaistic simulated environment and compare it to a metric map-based approach based on popular metric mapping method RTAB-MAP. The experimental results show that our topological approach significantly outperforms the metric one in terms of performance, keeping proper navigational efficiency.

While large language models (LLMs) have recently demonstrated strong potential in solving planning problems, there is a trade-off between flexibility and complexity. LLMs, as zero-shot planners themselves, are still not capable of directly generating valid plans for complex planning problems such as multi-constraint or long-horizon tasks. On the other hand, many frameworks aiming to solve complex planning problems often rely on task-specific preparatory efforts, such as task-specific in-context examples and pre-defined critics/verifiers, which limits their cross-task generalization capability. In this paper, we tackle these challenges by observing that the core of many planning problems lies in optimization problems: searching for the optimal solution (best plan) with goals subject to constraints (preconditions and effects of decisions). With LLMs' commonsense, reasoning, and programming capabilities, this opens up the possibilities of a universal LLM-based approach to planning problems. Inspired by this observation, we propose LLMFP, a general-purpose framework that leverages LLMs to capture key information from planning problems and formally formulate and solve them as optimization problems from scratch, with no task-specific examples needed. We apply LLMFP to 9 planning problems, ranging from multi-constraint decision making to multi-step planning problems, and demonstrate that LLMFP achieves on average 83.7% and 86.8% optimal rate across 9 tasks for GPT-4o and Claude 3.5 Sonnet, significantly outperforming the best baseline (direct planning with OpenAI o1-preview) with 37.6% and 40.7% improvements. We also validate components of LLMFP with ablation experiments and analyzed the underlying success and failure reasons.

Offline reinforcement learning (RL) is a powerful approach for data-driven decision-making and control. Compared to model-free methods, offline model-based reinforcement learning (MBRL) explicitly learns world models from a static dataset and uses them as surrogate simulators, improving the data efficiency and enabling the learned policy to potentially generalize beyond the dataset support. However, there could be various MDPs that behave identically on the offline dataset and so dealing with the uncertainty about the true MDP can be challenging. In this paper, we propose modeling offline MBRL as a Bayes Adaptive Markov Decision Process (BAMDP), which is a principled framework for addressing model uncertainty. We further introduce a novel Bayes Adaptive Monte-Carlo planning algorithm capable of solving BAMDPs in continuous state and action spaces with stochastic transitions. This planning process is based on Monte Carlo Tree Search and can be integrated into offline MBRL as a policy improvement operator in policy iteration. Our ``RL + Search" framework follows in the footsteps of superhuman AIs like AlphaZero, improving on current offline MBRL methods by incorporating more computation input. The proposed algorithm significantly outperforms state-of-the-art model-based and model-free offline RL methods on twelve D4RL MuJoCo benchmark tasks and three target tracking tasks in a challenging, stochastic tokamak control simulator.

Robots need robust and flexible vision systems to perceive and reason about their environments beyond geometry. Most of such systems build upon deep learning approaches. As autonomous robots are commonly deployed in initially unknown environments, pre-training on static datasets cannot always capture the variety of domains and limits the robot's vision performance during missions. Recently, self-supervised as well as fully supervised active learning methods emerged to improve robotic vision. These approaches rely on large in-domain pre-training datasets or require substantial human labelling effort. To address these issues, we present a recent adaptive planning framework for efficient training data collection to substantially reduce human labelling requirements in semantic terrain monitoring missions. To this end, we combine high-quality human labels with automatically generated pseudo labels. Experimental results show that the framework reaches segmentation performance close to fully supervised approaches with drastically reduced human labelling effort while outperforming purely self-supervised approaches. We discuss the advantages and limitations of current methods and outline valuable future research avenues towards more robust and flexible robotic vision systems in unknown environments.

In human cognition theory, human thinking is governed by two systems: the fast and intuitive System 1 and the slower but more deliberative System 2. Recent studies have shown that incorporating System 2 process into Transformers including large language models (LLMs), significantly enhances their reasoning capabilities. Nevertheless, models that purely resemble System 2 thinking require substantially higher computational costs and are much slower to respond. To address this challenge, we present Dualformer, a single Transformer model that seamlessly integrates both the fast and slow reasoning modes. Dualformer is obtained by training on data with randomized reasoning traces, where different parts of the traces are dropped during training. The dropping strategies are specifically tailored according to the trace structure, analogous to analyzing our thinking process and creating shortcuts with patterns. At inference time, our model can be configured to output only the solutions (fast mode) or both the reasoning chain and the final solution (slow mode), or automatically decide which mode to engage (auto mode). In all cases, Dualformer outperforms the corresponding baseline models in both performance and computational efficiency: (1) in slow mode, Dualformer optimally solves unseen 30 x 30 maze navigation tasks 97.6% of the time, surpassing the Searchformer (trained on data with complete reasoning traces) baseline performance of 93.3%, while only using 45.5% fewer reasoning steps; (2) in fast mode, Dualformer completes those tasks with an 80% optimal rate, significantly outperforming the Solution-Only model (trained on solution-only data), which has an optimal rate of only 30%. For math problems, our techniques have also achieved improved performance with LLM fine-tuning, showing its generalization beyond task-specific models.

Service robots that work alongside humans in a shared environment need a navigation system that takes into account not only physical safety but also social norms for mutual cooperation. In this paper, we introduce a motion planning system that includes human states such as positions and velocities and their personal space for social-aware navigation. The system first extracts human positions from the LiDAR and the RGB-D camera. It then uses the Kalman filter to fuse that information for human state estimation. An asymmetric Gaussian function is then employed to model human personal space based on their states. This model is used as the input to the dynamic window approach algorithm to generate trajectories for the robot. Experiments show that the robot is able to navigate alongside humans in a dynamic environment while respecting their physical and psychological comfort.

We consider the problem of synthesizing interpretable models that recognize the behaviour of an agent compared to other agents, on a whole set of similar planning tasks expressed in PDDL. Our approach consists in learning logical formulas, from a small set of examples that show how an agent solved small planning instances. These formulas are expressed in a version of First-Order Temporal Logic (FTL) tailored to our planning formalism. Such formulas are human-readable, serve as (partial) descriptions of an agent's policy, and generalize to unseen instances. We show that learning such formulas is computationally intractable, as it is an NP-hard problem. As such, we propose to learn these behaviour classifiers through a topology-guided compilation to MaxSAT, which allows us to generate a wide range of different formulas. Experiments show that interesting and accurate formulas can be learned in reasonable time.