A new round of experiments is complete. Our Multi-Hipothesys Monte Carlo localization algorithm, fully integrated with
#Nav2
, works perfectly at indoors/outdoors non-planar environments.
#science
#robotics
A new method using a Neural Object Field achieves 6-DoF tracking and 3D reconstruction of unknown, rigid objects from RGBD video sequences. It operates effectively even with occlusions, pose changes, and texture-less surfaces.
BundleSDF(CVPR 2023)
MPCC
Simulation environments in C++ and Matlab of the Model Predictive Contouring Controller (MPCC) for Autonomous Racing developed by the Automatic Control Lab (IfA) at ETH Zurich
mathematical_robotics
Optimization for Robotics
point_cloud_matching
Matching of 3D point clouds using rigid transformation by Lie groups, with optimization performed using Ceres Solver.
"Optimal Motion Generation-tools" is a Python software that simplifies motion planning. It's designed for both single-agent and multi-agent systems, and aims to make research on this topic more accessible.
omg-tools
viser
viser is a library for interactive 3D visualization + Python, inspired by tools like Pangolin, rviz, meshcat, and Gradio. It's designed to support applications in 3D vision and robotics.
Point cloud visualization
NeRF-RPN, a new object detection framework, operates directly on NeRF models and leverages multi-scale 3D neural volumetric features in a voxel representation to regress 3D bounding boxes without rendering.[CVPR 2023]
Traditional LiDAR SLAM yields sparse maps. A new CPU-only system generates and localizes dense mesh maps in real-time using Gaussian process reconstruction, running at 40Hz with enhanced accuracy.
SLAMesh
Mip-Splatting: Alias-free 3D Gaussian Splatting
> TL;DR: We introduce a 3D smoothing filter and a 2D Mip filter for 3D Gaussian Splatting (3DGS), eliminating multiple artifacts and achieving alias-free renderings.
CropCraft
A Procedural World Generator for Robotics Simulation of Agricultural Tasks
> CropCraft is a python script that allows to generate 3D models of crop fields specialized for real-time simulation of robotics application.
The paper presents OccNet, a novel pipeline for 3D occupancy reconstruction that improves driving tasks like detection, segmentation, and planning. Using a general occupancy embedding, it notably reduces collision rates in motion planning.
[AAAI 2024] Official repository for UCMCTrack: Multi-Object Tracking with Uniform Camera Motion Compensation. UCMCTrack ranks first on MOT17 without using any appearance cues.
Tracking Vehicles with Moving Camera
mppi_numba
A GPU implementation of Model Predictive Path Integral (MPPI) control that uses a probabilistic traversability model for planning risk-aware trajectories.
Automatic-Parking
Python implementation of an automatic parallel parking system in a virtual environment, including path planning, path tracking, and parallel parking
Detector-Free Structure from Motion
>TL;DR: We propose a detector-free structure from motion framework that eliminates the requirement of keypoint detection and can recover poses even on challenging texture-poor scene
Direct LiDAR Odometry:
Fast Localization with Dense Point Clouds
A lightweight and computationally-efficient frontend LiDAR odometry solution with consistent and accurate localization.
The paper outlines a new method for large-scale 3D reconstruction with LiDAR data. It utilizes an octree-based structure and a shallow neural network for data processing, and an incremental mapping system for continual learning.(ICRA' 23)
> Neural A* is a novel data-driven search-based planner that consists of a trainable encoder and a differentiable version of A* search algorithm called differentiable A* module.
Official implementation of "Path Planning using Neural A* Search"
Recognize Anything
The Recognize Anything Model (RAM) can recognize any common category with high accuracy.
When combined with localization models (Grounded-SAM), RAM forms a strong and general pipeline for visual semantic analysis.
TinyMPCTh
Pytorch Implementation of TinyMPC, a lightweight ADMM-based mpc solver. TinyMPC is division-free and requires no matrix factorization, which makes it robust and efficient.
Point-SLAM introduces a technique for RGBD using a point-based scene representation. Differing from traditional grid-based methods, this model adjusts anchor point density based on input information density, enhancing efficiency and accuracy(ICCV'23)
SelfOcc: Self-Supervised Vision-Based 3D Occupancy Prediction(CVPR 2024)
> SelfOcc empowers 3D autonomous driving world models (e.g., OccWorld) with scalable 3D representations, paving the way for interpretable end-to-end large driving models.
Trained…
The Topological Semantic Graph Memory framework enables robots to efficiently navigate unknown environments using landmark-based graph memory collected via an RGB-D camera, outperforming existing methods and demonstrating success in real-world scenarios
sr_livo
A LiDAR-inertial-visual odometry and mapping system based on the sweep reconstruction method
> SR-LIVO is designed based on the framework of R3Live.
> We employ the sweep reconstruction method to align reconstructed sweeps with image timestamps
Fuzzy Metaballs+ use 3D Gaussians for differentiable rendering, integrating optical flow and producing watertight meshes.
These techniques ensure speedy, reliable reconstructions on both GPU and CPU.
google-research/ibc
Implicit Behavioral Cloning
Implicit BC policy on a precise, 1-millimeter-tolerance
slide-then-insert task: push a block across a table, then slide it into a slot.
curobo
CUDA Accelerated Robot Library
> CuRobo is a CUDA accelerated library containing a suite of robotics algorithms that run significantly faster than existing implementations leveraging parallel compute.
The software by zhm-real is a collection of key algorithms for autonomous vehicle planning and tracking, featuring Hybrid A* Planner, Frenet Optimal Trajectory, and multiple controllers.
Hybrid A* Planner
Khronos
Spatio-Temporal Metric-Semantic SLAM
> Khronos is a unified approach that can reason about short-term dynamics and long-term changes when performing online metric-semantic simultaneous mapping and localization (SLAM) in dynamic environments
LMDrive
Closed-Loop End-to-End Driving with LLM
An end-to-end, closed-loop, language-based autonomous driving framework, which interacts with the dynamic environment via multi-modal multi-view sensor data and natural language instructions
GraphBasedLocalTrajectoryPlanner for autonomous race vehicles employs a multilayer graph framework, providing optimal, feasible actions, and has demonstrated success in real-world applications at speeds above 200kph during Roborace Season Alpha.
python_simple_mppi
Python implementation of MPPI(Model Predictive Path-Integral) controller to understand the basic idea. Mandatory dependencies are numpy and matplotlib only.
SLAMBOX is designed for use metod SLAM in education, experiments, research and development by Node-based user interface. This is a box with tools that you can quickly and conveniently experiment with separate SLAM nodes
libRSF is a robust C++ library for sensor fusion with sliding window filtering, predefined cost functions, and robust error models, leveraging Ceres Solver
libRSF
A robust sensor fusion library for online localization.
RGB-D画像のマッチング精度を向上させるための画像特徴抽出と融合手法を提案。
SIFT、SURF、ORB特徴量とPFH、FPFH特徴量を組み合わせた特徴量を使用し、誤マッチングを減少させている。
GlueStick: Robust Image Matching by Sticking Points and Lines Together