Skip to content
#Perception and localization
Mar 18Tue
  1. FieldAI39

    FieldAI Research Institute (FAIRI) studies embodied question answering in dynamic environments based on the NVIDIA Omniverse platform

    Field AI Research Institute (FAIRI) is collaborating with NVIDIA to study robotic embodied question answering (EQ&A) in dynamic environments. The team uses the NVIDIA Omniverse platform, Isaac Sim, Omniverse Replicator, and the USD Search API to build physically accurate 3D office scenes, simulate changes in the environment over time, and plans to open-source this long-term embodied question answering dataset.

Feb 12Wed
  1. Deep Robotics/云深处科技30

    XGRIDS and DEEP Robotics Reach Strategic Partnership to Advance Integration of 3D Spatial Computing and Embodied Intelligence

    XGRIDS and DEEP Robotics recently reached a strategic partnership, combining XGRIDS' high-precision 3D scanning and modeling equipment with the environmental interaction capabilities of DEEP Robotics' intelligent robot dogs to create an intelligent modeling solution. The system targets scenarios such as tunnel and underground space exploration, emergency rescue, and building and campus surveying, supports 4G, 5G, and ad hoc network modes, and can transmit data in real time in areas without signal coverage.

Jul 8Mon
  1. 泡泡机器人SLAM · 微信公众号48

    [PaoPao Point Cloud Space-Time] Confidence-Rich Mutual Information for Information-Theoretic Mapping

    A RAL 2021 paper proposes Confidence-Rich Mutual Information (CRMI), which computes information gain based on continuous probability distributions of occupancy grid cells and a beam sensor causal model (SCM), avoiding the shortcomings of the traditional inverse sensor model (ISM). Simulations and experiments show that CRMI-driven robots explore unexplored and ambiguous regions more finely, with quadratic complexity close to traditional OGMI, and can run online.

Mar 19Tue
  1. 泡泡机器人SLAM · 微信公众号45

    DiSCo-SLAM: Distributed Scan Context-Enabled Multi-Robot LiDAR SLAM with Two-Stage Global-Local Graph Optimization

    DiSCo-SLAM is the first distributed framework to use lightweight Scan Context descriptors for multi-robot SLAM, targeting 3D LiDAR observations and supporting efficient data exchange between robots. The framework includes two-stage global-local optimization and can provide stable localization results under unknown initial conditions for inter-robot loop closure search. Compared with the distributed Gauss-Seidel (DGS) method on multiple multi-robot datasets, its accuracy, stability, and data efficiency were quantitatively verified.

Feb 19Mon
  1. 泡泡机器人SLAM · 微信公众号41

    【Bubble One Minute】Efficient-Communication, Robust and Accurate Multi-Agent Collaborative SLAM Based on Smartphones

    RAMA-SLAM proposes a multi-agent collaborative visual-inertial SLAM system based on a centralized client-server (CS) architecture, with clients running on smartphones, achieving experience sharing through enhanced geometric verification, error-aware weighted reprojection optimization, and a flexible map fusion strategy. The system was tested on the EuRoC dataset and in real environments, where results outperformed existing methods, and it demonstrated the stability of collaborative SLAM in a real environment with 7.55 Mbps bandwidth.

Jan 26Fri
  1. 泡泡机器人SLAM · 微信公众号55

    M2DGR: A Multi-Sensor, Multi-Scenario SLAM Dataset for Ground Robots

    M2DGR is a large-scale SLAM dataset collected by a ground robot, comprising 36 indoor and outdoor scene sequences and about 1TB of data. Its sensor suite includes six fisheye cameras, a skyward-facing RGB camera, an infrared camera, an event camera, a visual-inertial sensor, an IMU, a lidar, a consumer-grade GNSS receiver, and a GNSS-IMU navigation system with RTK signals, all calibrated and synchronized and recorded in sync.

Jan 23Tue
  1. 泡泡机器人SLAM · 微信公众号47

    Self-Supervised Depth and 3D Motion Estimation from Monocular Thermal Video Based on Multispectral Consistency Loss

    A study published at ICRA 2022 proposed a self-supervised learning method that simultaneously estimates depth and ego-motion from monocular thermal video. The method uses multispectral consistency consisting of a temperature consistency loss and a photometric consistency loss, and designs a differentiable forward warping module that transforms the thermal camera coordinate system into a visible-light camera, thereby providing a complementary self-supervised signal.

Jan 9Tue
  1. 泡泡机器人SLAM · 微信公众号41

    Robust Object Detection Under Adverse Conditions via Event Camera and RGB Camera Fusion

    An ICRA 2022 paper proposes a redundant sensor fusion model of event cameras and RGB cameras for robust object detection under image corruption and different weather conditions. The method takes voxel grid representations of events as input and processes frames and events separately through two parallel feature extractor networks. After training and evaluation on the public DSEC dataset, the fusion method improved robustness to common image corruptions by more than 30% compared with frame-only detection, and outperformed event-only detection.

Dec 25Mon
  1. 泡泡机器人SLAM · 微信公众号57

    NeRF-LOAM: Neural Implicit Representation for Large-Scale Incremental LiDAR Odometry and Mapping

    Bubble Turing Think Tank compiled and introduced the arXiv 2023 paper NeRF-LOAM, which the authors describe as the first neural implicit odometry and mapping system for large-scale environments using LiDAR data. It consists of three parts—Neural Tracking, Neural Mapping, and Mesh Reconstruction—all based on the Neural SDF they propose.

Dec 13Wed
Dec 11Mon
  1. 泡泡机器人SLAM · 微信公众号43

    [Bubble One Minute] Interval-based visual-inertial LiDAR SLAM using anchor poses (i-VIL SLAM)

    i-VIL SLAM proposes a visual-inertial LiDAR SLAM method that assumes only bounded sensor errors, propagates errors to the map and trajectory through interval analysis, and stabilizes uncertainty accumulation using anchor poses derived from GNSS/INS data. Experiments show that in the worst case, the 99.9% confidence ellipsoid derived from marginal covariance in classical SLAM graph optimization contains less than 64% of ground truth, whereas the pose interval bounds of this method contain ground truth for more than 96% of frames.

Nov 17Fri
  1. OFweek Robotics56

    Leishen Intelligent secures over 100 million yuan in funding to enter 3D SLAM unmanned forklifts

    LiDAR company Leishen Intelligent received over 100 million yuan in funding in October, co-led by Jinan Tongxin Future Industrial Investment and Chongqing Nanbu Fund, and will increase investment in 3D SLAM unmanned forklift R&D and base construction. Its 3D SLAM solution deploys faster and adapts to environments better than 2D SLAM, with a single LiDAR capable of mapping up to one million square meters; the product series has already been deployed in over 150 projects across multiple industries.

May 23Tue
  1. OFweek Robotics37

    Interactive robot "Stargazer" arrives: it can be your personal photographer

    An interactive robot called "Stargazer," developed by a team at the University of Toronto, uses a seven-motor robotic arm to drive a camera and autonomously track its subject, acting as a personal photographer for video creators. It connects to ChatGPT-3 and Microsoft Azure Speech-to-Text speech recognition and can understand instructions such as high-angle shots and tight framing. It is currently used mainly for testing in teachers' instructional video recording. The team's next step is to improve instruction recognition accuracy and extend the technology to drones or wheeled robots.

Dec 16Fri
  1. OFweek Robotics45

    Viewtrix and CRRC Digital deliver industry's first large-part automated warehouse gantry handling and sorting project, reaching full production

    Viewtrix Robotics delivered the industry's first large-part automated warehouse gantry handling and sorting project at the smart factory of a major construction machinery company. After half a year, the project passed acceptance at full production, raising the handling and sorting efficiency of multi-ton large parts to 120 seconds per piece.

Oct 12Wed
Sep 21Wed
  1. 机器人产业网 · 微信公众号25

    Brief: CAS brain-computer interface decoding advance, world's first flying bicycle debuts, liquid robot can split, and more

    The Tang Fengzhen research group at the Shenyang Institute of Automation, Chinese Academy of Sciences, proposed an EEG signal decoding method based on the log-Euclidean metric Riemannian geometry. Its recognition speed is about 4 times faster than methods based on the affine-invariant Riemannian metric, and its recognition accuracy on the BCI Competition IV-2a dataset outperforms the competition's first-place entry.

Sep 30Thu
  1. OFweek Robotics39

    All-Scenario, Digital: What Makes UDI Service Robots Different?

    UDI Technology spans multiple scenarios including entertainment and dining, hotels, and outdoor delivery. It has achieved mass production and shipment of products including the fifth-generation You Xiaodi, You Xiaomei, and third-generation You Xiaoge, and has partnered with more than 8,000 B-end customers in hotel scenarios cumulatively. Its self-developed SLAM localization and mapping, differential chassis, and motion control technologies use the NVIDIA Xavier computing platform. Sales revenue in the first half of 2021 grew 9-fold year on year. The outdoor delivery business is expected to carry out large-scale commercial deployment testing by the end of this year.

Sep 2Thu
  1. OFweek Robotics44

    Tips for Getting the Most Out of the Roborock G10 Self-Cleaning Robot Vacuum and Mop

    The Roborock G10 self-cleaning robot vacuum and mop combines sweeping, mopping, and vacuuming. For mopping, it uses a high-frequency vibration motor of up to 3,000 rpm and a flat mop pad; the pad can automatically lift more than 5 mm off the floor, and an ultrasonic carpet-recognition sensor lets it vacuum only, without mopping, on carpets. It is also the industry's first robot vacuum and mop with automatic water-tank refilling. Its self-cleaning base washes the mop pad through a "wash, brush, scrape" process, and the company claims a laboratory cleaning rate of up to 90%. The app offers four mop-washing modes: Standard, Smart, Manual, and Manual Dock Return.

Aug 30Mon
  1. OFweek Robotics50

    Roborock G10 self-cleaning robot vacuum opens pre-sale: automatic mop washing, water tank refill, first-sale price 3999 yuan

    The Roborock G10 self-cleaning robot vacuum is now available for pre-sale on the Roborock store, with a first-sale price of 3999 yuan. The model supports high-frequency sonic mopping at up to 3000 vibrations per minute, automatic mop washing and scraping (95% cleaning rate), and automatic water tank refill; the mop can lift intelligently to handle carpets and overcome obstacles. The navigation algorithm has been upgraded to RR mason 8.0, which can intelligently recognize multi-floor maps.

Apr 21Wed
  1. OFweek Robotics19

    How Yunson Intelligent Builds Full-Scenario Intelligent Service Robots

    At the 3rd China International Service Robot Innovation and Development Conference, Shi Wenhua, Vice President of Technology at Yunson Intelligent, presented a full-scenario intelligent service solution centered on AI service robots. As a wholly owned subsidiary of Red Star Macalline Holding, the company builds smart mall, smart building, and smart community solutions, supported by an integrated scheduling and management platform. It holds more than 100 related invention patents, and its products have been applied in scenarios such as mall shopping guidance, exhibition hall tours, and government services.

Apr 9Fri
  1. OFweek Robotics27

    Rubotics' Qiu Qiang: Using an "intelligent hand-eye unit" combining motion planning and machine vision to address the intelligence and cost pain points of industrial robots

    Qiu Qiang, CTO of Shenzhen Rubotics, said the company uses machine vision and robot motion planning technology to build an intelligent hand-eye unit, targeting the two major pain points of industrial robots: insufficient intelligence and high application costs. Its bathroom polishing unit can use 3D vision to detect the shape of a washbasin and automatically plan the working trajectory, shortening debugging time; the intelligent hand-eye unit can free 1 to 2 workers from hazardous environments. On February 1 this year, Rubotics signed a cooperation agreement with Japan's RBB company, beginning its "going global" push.