Blog Posts

  • Lidar odometry smoothing using ES EKF and KissICP for Ouster sensors with IMUs

    In this post I've explored the possibility of improving the KissICP trajectory output for Ouster Lidars using the sensor's embedded IMUs. On some enviromnents ES EKF filtering together with KissICP made the 10% decrease in the ATE for high resolution 128 beam sensors but on other environments with lower resolution 32 beam sensors it showed 50% decrease in ATE compared to ground truth. ES EKF filter formulation and experiments described with references, source code and CLI tools for visualizations that may be used to replicate and test approaches on any Ouster's Lidar pcap/bag raw data recordings.

  • P(oint) (e)Tudes: Lidar odometry, SLAM and visualization tools

    Spending more than 4 years with lidar sensors at Ouster I couldn't stop wanting more, diverse and robust tools to work with 3D data. With the Point eTudes playground I'm extending capabilities to work with lidar sensors. First tool is a "flyby" over the registered 3D point cloud map. Second tool is a "viz" for visualizing raw sensor ROS bags, for example Newer College Dataset bags.

  • 3D Reconstruction using Structure from Motion (SfM) pipeline with OpenGL visualization on C++

    In this project, I explore the traditional SfM pipeline and build sparse 3D reconstruction from the Apolloscape ZPark sample dataset with simultaneous OpenGL visualization. This article is the summary where I will touch things from keypoints detection and matching, to epipolar constraints and bundle adjustment optimization using Ceres solver.

  • PoseNet implementation for self-driving car localization using Pytorch on Apolloscape dataset

    In this post, I explore the Apolloscape dataset for self-localization task with Pytorch implementation of PoseNet architecture with the automatic learning of weighting params for rotation and translation components in a combined loss. The self-localization task is a part of ECCV 2018 Workshop on Vision-Based Navigation.

  • Robotics Projects: Navigation, Mapping, Control, Deep RL and others

    The list of robotics projects that I've completed during my study at Udacity's Robotics Nanodegree program with short problem descriptions, results and links to the relevant materials to learn more about topics like perception (2D/3D), control, navigation, mapping, planning, deep reinforcement learning and ROS/Gazebo integrations.

  • MS Build 2018 Notes: Mixed Reality and Smart Enterprises

    Notes from Microsoft Build 2018 events and thoughts about where AR/MR/VR are moving along with smart enterprises. Grow of the real-time and edge computing, cognitive services and hardware + software + cloud bundles.

  • Train Fingers Classification Model Using NVidia DIGITS

    Train LeNet network to recognize a number of fingers using NVidia DIGITS platform with the subsequent deployment of the model to the mobile robot running on Jetson TX2.

  • Low Earth Orbit (LEO) and delta-V requirements

    What speed and altitude do you need to successfully reach and stay some time in an Earth orbit.

  • Links to my 7 older posts (2015)

    Collection of links to select posts that I wrote before Dec 2017.

subscribe via RSS