Unitree G1
Bipedal humanoid robot by Unitree Robotics, designed as a compact research, development, and developer platform.
Applications
- Research
- Home Assistance
Perception · Perception & Vision Software
1.2.0·University of Bonn — Photogrammetry & Robotics Lab (Stachniss Lab)
KISS-ICP (Keep It Small and Simple Iterative Closest Point) is an open-source LiDAR odometry system designed at the Photogrammetry & Robotics Lab at the University of Bonn under prof. Cyrill Stachniss. The main idea: most LiDAR-SLAM systems have become overly complicated — KISS-ICP proves that a minimalist pipeline can achieve state-of-the-art accuracy on public benchmarks (KITTI, MulRan, NewerCollege) without deep learning and without fancy descriptors.
The algorithm: (1) point cloud deskewing based on motion prediction from the previous step (constant velocity model), (2) cloud downsampling, (3) point-to-point ICP with robust kernel (Geman-McClure) against the map, (4) map update as a voxel grid with adaptive threshold. No external IMU, no loop closure, no ML — pure well-tuned ICP.
The implementation is in C++ with Python bindings, uses Eigen and Sophus for algebra, TBB for parallelization. The entire pipeline fits in ~3000 lines of code, making it a pedagogical and practical reference implementation of ICP. Real-world performance: ~50 Hz on a typical Intel Core i7 for a Velodyne VLP-16 scan (~30k points), drift translation ~1-2% of distance on KITTI.
KISS-ICP is gaining popularity as a foundation for more advanced systems: users often combine it with loop closure (e.g. Scan Context, OverlapNet) or IMU fusion (e.g. LIO-SAM, FAST-LIO) to obtain a full SLAM stack. It is also used as a baseline benchmark in scientific papers — if your system does not beat KISS-ICP, it is probably not state of the art.
MIT licensed, ROS 2 wrapper from the official repo (kiss-icp-ros2), package available via pip (kiss-icp). Actively developed 2022-2026, regular releases, small but strong academic community.
An API Library is a software package that exposes programmatic interfaces for communicating with a device, service, or system. In robotics it typically forms a lightweight integration layer built on top of the manufacturer's official API or an open-source project, abstracting low-level protocol details and providing language-native bindings (Python, C++, Java, etc.).
A Perception Stack encompasses the software layers that process data from cameras, LiDARs, IMUs, microphones, and other sensors in order to recognise the surrounding environment, perform localisation, detect and track objects, and interpret the scene. It is typically the first processing stage in an autonomous robot's data pipeline, feeding its outputs to planning and control stacks.
A family of open-source robotic perception libraries: Visual SLAM (ORB-SLAM, RTAB-Map), computer vision (OpenCV), point cloud processing (PCL, Open3D), detection and segmentation (YOLO, Detectron2).
Academia: KISS-ICP used as baseline in hundreds of LiDAR-SLAM papers (IROS, ICRA, RA-L, IJRR) — comparison foundation for new methods. Boston Dynamics Spot research community — KISS-ICP used in research tests of quadruped gait and complex environment mapping (PRBonn integration). Autonomous driving: lightweight deployments in prototype autonomous shuttles (Karlsruhe, Bonn, Aachen), where minimal code size and absence of external ML make KISS-ICP attractive for certification. NewerCollege Dataset (Oxford) — KISS-ICP used as reference LiDAR-only odometry baseline. Drones and UAV: integration with FAST-LIO/LIO-SAM as odometry warm-start (PRBonn forks). Mobile robotics community (AGV/AMR) — KISS-ICP increasingly chosen over Cartographer for new projects due to its simplicity and predictability.
GitHub: PRBonn/kiss-icp — ~2 000 stars, ~300 forks, ~25 contributors. Active issues, regular PRs from the academic community. Citations: the original paper Vizzo, Guadagnino, Mersch, Wiesmann, Behley, Stachniss 'KISS-ICP: In Defense of Point-to-Point ICP — Simple, Accurate, and Robust Registration if Done the Right Way' (IEEE RA-L 2023) — >500 citations in less than 3 years. Community: strong presence in the LiDAR-SLAM environment (PRBonn lab seminars, ROS Discourse, robotics-SLAM Discord). Small but expert — mainly PhD students and robotics engineers from autonomous driving companies. Maintenance: active — Ignacio Vizzo (main author) leads development, regular releases ~every 3-6 months. No commercial sponsor — academic project funded by DFG and ERC grants of the Stachniss Lab.
Bipedal humanoid robot by Unitree Robotics, designed as a compact research, development, and developer platform.
Full-size general-purpose humanoid (~180 cm, ~47 kg). Bipedal, 5 DOF per leg, 3.3 m/s walking speed, 360° perception via 3D LiDAR + depth camera, Unitree M107 PMSM joint motors (~360 N·m peak knee torque).
Boston Dynamics bipedal humanoid robot. The fully electric generation unveiled in 2024 succeeds the hydraulic Atlas that was retired after more than a decade of research.
Tutor Intelligence industrial robot designed for palletizing, depalletizing and case-picking in warehouses and factories, offered as Robotics-as-a-Service from $14/hr.
Compact, high-dynamics bipedal humanoid by MagicLab. 140 cm, 40 kg, 24–50 DOF, walking speed up to 2.5 m/s. Unveiled on 8 July 2025 with martial arts and acrobatic demos.
Ubuntu 24.04 LTS 'Noble Numbat' — supported until April 2029. The host for ROS 2 Jazzy.
Requires Eigen 3.3+, Sophus, TBB. Python bindings need pybind11. ROS 2 wrapper (separate package) requires Humble or newer.
License family: Permissive
Support for Sequential Scan Matching (SSM), robustness improvements for low-frequency LiDAR (5 Hz).
New adaptive threshold for voxel grid, better Livox spinning LiDAR support.
Stable 1.0 API, IEEE RA-L publication 'KISS-ICP: In Defense of Point-to-Point ICP'.
Python bindings via pybind11, CI integration with ROS 2.
ROS 2 wrapper (kiss-icp-ros2) as separate repo, adaptive threshold refactor.
Introduction of Geman-McClure robust kernel, multi-threaded ICP via TBB.
First public release — minimalist point-to-point ICP pipeline.
