docs

Welcome to the docs for tinyNav.

Overview

TinyNav is a compact, efficient, and modular navigation framework designed primarily for use with Intel RealSense depth cameras and NVIDIA Jetson platforms. It provides essential capabilities for stereo-based visual odometry, mapping, and path planning in real-time robotics applications.

Module introduce

Stereo Visual Odometry Module

The odometry module is expected to use pairs of images for relative pose estimation through feature extraction and feature matching procedures.

Function	Description
Depth Estimation	Computes dense depth maps from stereo disparity images using Semi-Global Block Matching (SGBM), enabling 3D reconstruction and localization.
Feature Extraction & Matching	Detects keypoints with SuperPoint and matches them using LightGlue.
Pose Estimation	Estimates relative camera pose via 3D-2D correspondences using PnP methods.
Odometry Node Publishing(Optional)	Publishes estimated pose as nav_msgs/Odometry and broadcasts the corresponding TF.

Planning Module

The planning module is responsible for generating optimal and feasible control commands for the robot, utilizing information from local perception.

Function	Description
Local Dynamic Mapping	Generates a local occupancy grid map from accumulated depth frames and robot poses in real-time.
Trajectory Simulation & Scoring	Uses Dynamic Window Approach (DWA) to simulate multiple candidate trajectories, scoring them based on goal alignment, obstacle avoidance.
Optimal Command Selection	Selects the best trajectory and computes the corresponding velocity commands for execution.
Planned Command Node Publishing(optional)	Publishes planned command as geometry_msgs/Twist and local map as nav_msgs/OccupancyGrid.

Mapping Module

The mapping module is responsible for providing the robot’s global position along with static map information.

Function	Description
Global Feature Extraction	Extracts global image descriptors using DINO V2 embeddings for scene recognition, loop detection, and global similarity assessment.
Local Bundle Adjustment	Performs local optimization of camera poses and 3D landmark positions using a Bundle Adjustment solver while respecting relative pose constraints.
Loop Closure	Identifies potential loop closure candidates via DINO feature similarity, refines alignment through pose graph optimization, and updates the map via BA.
Map Node Publishing (Optional)	Publishes the path as nav_msgs/Path.
Global Planning (Optional)	Can be extended to include global path planning using cost maps and A* method.

Set up for Unitree Go2

Connect your computer or Jetson to the Unitree Go2 and ensure they are on the same network segment (192.168.123.x).

Step 1: Find the Network Interface Name

Run the following command to list all network interfaces:

ip addr

Example output:

1: lo: <LOOPBACK,UP,LOWER_UP> mtu 65536 qdisc noqueue state UNKNOWN group default qlen 1000
    link/loopback 00:00:00:00:00:00 brd 00:00:00:00:00:00
    inet 127.0.0.1/8 scope host lo
       valid_lft forever preferred_lft forever
    inet6 ::1/128 scope host 
       valid_lft forever preferred_lft forever
2: wlP1p1s0: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc mq state UP group default qlen 1000
    link/ether 48:8f:4b:62:b4:4c brd ff:ff:ff:ff:ff:ff
    inet 192.168.19.115/24 brd 192.168.19.255 scope global dynamic noprefixroute wlP1p1s0
       valid_lft 12597sec preferred_lft 12597sec
    inet6 fe80::edfc:6e55:62ab:bfd3/64 scope link noprefixroute 
       valid_lft forever preferred_lft forever
3: can0: <NOARP,ECHO> mtu 16 qdisc noop state DOWN group default qlen 10
    link/can 
4: eno1: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc mq state UP group default qlen 1000
    link/ether 3c:6d:66:b2:b3:f1 brd ff:ff:ff:ff:ff:ff
    inet 192.168.123.100/24 scope global eno1
       valid_lft forever preferred_lft forever

Note: In this example, eno1 is the network interface connected to the Go2 robot (IP: 192.168.123.100).

Step 2: Set a Static IP Address

Configure your network interface with a static IP address in the 192.168.123.x range:

sudo ip addr add 192.168.123.100/24 dev eno1
sudo ip link set eno1 up

Step 3: Verify Connection

Test the connection to the Go2 robot (default IP: 192.168.123.161):

ping 192.168.123.161

Additional Resources

For more detailed setup instructions, refer to the official Unitree documentation: https://support.unitree.com/home/en/developer/Quick_start