An open problem in mobile manipulation is how to represent objects and scenes in a unified manner so that robots can use both for navigation and manipulation. The latter requires capturing intricate geometry while understanding fine-grained semantics, whereas the former involves capturing the complexity inherent at an expansive physical scale. In this work, we present GeFF (Generalizable Feature Fields), a scene-level generalizable neural feature field that acts as a unified representation for both navigation and manipulation that performs in real-time. To do so, we treat generative novel view synthesis as a pre-training task, and then align the resulting rich scene priors with natural language via CLIP feature distillation. We demonstrate the effectiveness of this approach by deploying GeFF on a quadrupedal robot equipped with a manipulator. We quantitatively evaluate GeFF’s ability for open-vocabulary object-/part-level manipulation and show that GeFF outperforms point-based baselines in runtime and storage-accuracy trade-offs, with qualitative examples of semantics-aware navigation and articulated object manipulation.

With the rapid development of the internet of vehicles (IoV) system and intelligent applications, vehicles are generating increasingly computation-intensive tasks, which impose stringent demands on real-time performance, energy efficiency, and resource management. Recently, multi-agent deep reinforcement learning (MADRL) has emerged as a promising solution for task offloading in such dynamic environments, owing to its adaptability and low decision-making complexity. However, most existing approaches predominantly focus on resource constraints at the vehicle level, often neglecting the limited computing and storage capacities of edge servers, such as roadside units (RSUs), which results in inefficient offloading and coordination decisions. To address these challenges, we propose the Combinatorial Multi-Agent Actor-Critic with Coordination (COMA2C) algorithm. In this algorithm, vehicle agents dynamically generate task offloading requests based on link conditions and local resource availability, while RSU agents, constrained by limited resources, perform centralized decision-making through a combinatorial action selection mechanism. This enables global resource coordination and efficient task allocation. Simulation results under typical Vehicular Edge Computing (VEC) scenarios show that the proposed approach outperforms existing benchmarks in terms of task completion rate, task delay and energy consumption, thereby validating its effectiveness and scalability in real-world vehicular environments.

Autonomous robots frequently need to detect “interesting” scenes to decide on further exploration, or to decide which data to share for cooperation. These scenarios often require fast deployment with little or no training data. Prior work considers “interestingness” based on data from the same distribution. Instead, we propose to develop a method that automatically adapts online to the environment to report interesting scenes quickly. To address this problem, we develop a novel translation-invariant visual memory and design a three-stage architecture for long-term, short-term, and online learning, which enables the system to learn human-like experience, environmental knowledge, and online adaption, respectively. With this system, we achieve an average of 20% higher accuracy than the state-of-the-art unsupervised methods in a subterranean tunnel environment. We show comparable performance to supervised methods for robot exploration scenarios showing the efficacy of our approach. We expect that the presented method will play an important role in the robotic interestingness recognition exploration tasks.

We present a challenging dataset, the TartanAir, for robot navigation tasks and more. The data is collected in photo-realistic simulation environments with the presence of moving objects, changing light and various weather conditions. By collecting data in simulations, we are able to obtain multi-modal sensor data and precise ground truth labels such as the stereo RGB image, depth image, segmentation, optical flow, camera poses, and LiDAR point cloud. We set up large numbers of environments with various styles and scenes, covering challenging viewpoints and diverse motion patterns that are difficult to achieve by using physical data collection platforms. In order to enable data collection at such a large scale, we develop an automatic pipeline, including mapping, trajectory sampling, data processing, and data verification. We evaluate the impact of various factors on visual SLAM algorithms using our data. The results of state-of-the-art algorithms reveal that the visual SLAM problem is far from solved. Methods that show good performance on established datasets such as KITTI do not perform well in more difficult scenarios. Although we use the simulation, our goal is to push the limits of Visual SLAM algorithms in the real world by providing a challenging benchmark for testing new methods, while also using a large diverse training data for learning-based methods. Our dataset is available at http://theairlab.org/tartanair-dataset.

Three diode-capacitor-inductor networks which can be operated for stepping-up the input voltage are proposed. These networks differ one from the other by the number and position of their components. By inserting these networks into a general converter structure, three novel large DC gain non-isolated converters are generated. The single MOSFET switch in each of these converters is referenced to the ground, rendering the drive circuit relatively simple. The smaller size of capacitors in these converters can decrease the size of the overall converter, resulting in a higher power density. These converters can achieve considerably large DC voltage conversion ratios, what is necessary in many high step-up applications like those in conjunction with the environmental friendly sources of energy.

A new step-up basic switching cell formed by four passive elements and two diodes is proposed. It is inserted in a boost converter in order to obtain a new hybrid PWM non-isolated converter able to provide a large DC gain, as required as front-end to electrical grids supplied by solar or fuel cells. A family of converters can be created by inserting the defined switching structure in any step-up converter. Analysis, simulation and experimental data show the superiority of the proposed converter over other hybrid converters with similar complexity and the same voltage ratio: less voltage stresses on the switches.

A simple switched-capacitor step-up cell formed by two capacitors and two diodes is inserted in a boost converter in order to get a large dc line-to-load voltage ratio. The output capacitor and output diode of the boost converter are eliminated, their role being taken by the elements of the switching block. The dc gain of the new structure is the same or larger than that of other converters with two capacitors and one inductor. The voltage stress of the transistor and diodes is just the difference between the output and input voltages. In both switching stages, a direct transfer path of energy from line to load is created, increasing thus the energy conversion efficiency. The new hybrid converter represents another solution for the interface between the alternative sources of energy and their loads or grid. The simulation and experimental results confirm the theoretical analysis.