docs/modules/isaac/chapter2.md

metadata

sidebar_position: 3

Chapter 2: Isaac ROS and VSLAM Navigation

Learning Objectives

• Understand the Isaac ROS ecosystem and its components
• Learn about Visual Simultaneous Localization and Mapping (VSLAM)
• Implement VSLAM systems for humanoid robot navigation
• Integrate Isaac ROS perception packages with navigation systems
• Configure Nav2 for humanoid robot applications

Isaac ROS Introduction

Isaac ROS is NVIDIA's collection of hardware-accelerated, perception-focused packages designed for robotics applications. These packages leverage NVIDIA's GPUs to accelerate perception tasks, which is especially important for humanoid robots that require real-time processing of multiple sensor streams.

Key Isaac ROS Packages

1. Isaac ROS Image Pipeline: GPU-accelerated image processing
2. Isaac ROS Visual SLAM: GPU-accelerated visual SLAM
3. Isaac ROS Object Detection: Real-time object detection
4. Isaac ROS Apriltag: AprilTag detection and pose estimation
5. Isaac ROS Stereo Dense Reconstruction: 3D environment reconstruction

Advantages for Humanoid Robots

Isaac ROS provides specific benefits for humanoid robotics:

• GPU Acceleration: Critical for processing multiple sensors in real-time
• High-Performance SLAM: Essential for localization in complex environments
• Robust Perception: Important for safe navigation around humans
• Real-time Processing: Necessary for dynamic balance and control

Visual SLAM (VSLAM) Fundamentals

Visual SLAM (Simultaneous Localization and Mapping) combines visual data with odometry to create maps while tracking robot position within them. This is particularly valuable for humanoid robots operating in human environments where traditional LIDAR-based SLAM might be insufficient.

How VSLAM Works

1. Feature Detection: Identify distinctive points in visual data
2. Feature Matching: Match features between consecutive frames
3. Motion Estimation: Estimate camera motion based on feature movement
4. Mapping: Build a map of the environment using visual features
5. Optimization: Refine the map and trajectory estimates

VSLAM vs. Traditional SLAM

Visual SLAM offers advantages over traditional LIDAR-based approaches:

• Rich Information: Visual data contains more semantic information
• Lower Cost: No need for expensive LIDAR sensors
• Better for Indoor Environments: Works well in texture-rich environments
• Human-like Perception: More similar to human navigation

However, it also has challenges:

• Lighting Sensitivity: Performance degrades in poor lighting
• Dynamic Objects: Moving objects can cause tracking errors
• Computational Requirements: More processing power needed

Isaac ROS Visual SLAM Package

The Isaac ROS Visual SLAM package provides hardware-accelerated visual SLAM using NVIDIA GPUs.

Key Features

• GPU Acceleration: Utilizes CUDA cores for feature detection and matching
• Real-time Performance: Capable of processing video at high frame rates
• Robust Tracking: Handles viewpoint changes and lighting variations
• ROS 2 Compatibility: Integrates seamlessly with ROS 2 navigation stack

Installation

Isaac ROS packages are typically installed via Docker containers:

# Pull the Isaac ROS Docker container
docker pull nvcr.io/nvidia/isaac-ros:ros-humble-visualslam-cu11.8.0-22.12.2

# Run with GPU access
docker run --gpus all -it --rm \
  --network=host \
  --env="DISPLAY" \
  --env="QT_X11_NO_MITSHM=1" \
  --volume="/tmp/.X11-unix:/tmp/.X11-unix:rw" \
  nvcr.io/nvidia/isaac-ros:ros-humble-visualslam-cu11.8.0-22.12.2

Launching Isaac ROS Visual SLAM

<!-- visual_slam.launch.py -->
from launch import LaunchDescription
from launch_ros.actions import ComposableNodeContainer
from launch_ros.descriptions import ComposableNode

def generate_launch_description():
    container = ComposableNodeContainer(
        name='visual_slam_container',
        namespace='',
        package='rclcpp_components',
        executable='component_container_mt',
        composable_node_descriptions=[
            ComposableNode(
                package='isaac_ros_visual_slam',
                plugin='nvidia::isaac_ros::visual_slam::VisualSlamNode',
                name='visual_slam_node',
                parameters=[{
                    'enable_rectified_pose': True,
                    'map_frame': 'map',
                    'odom_frame': 'odom',
                    'base_frame': 'base_link',
                    'enable_fisheye_distortion': False,
                }],
                remappings=[
                    ('/visual_slam/image_raw', '/camera/image_rect'),
                    ('/visual_slam/camera_info', '/camera/camera_info'),
                ],
            ),
        ],
        output='screen',
    )

    return LaunchDescription([container])

Nav2 for Humanoid Robots

Nav2 is the standard navigation framework for ROS 2. While traditionally used for wheeled robots, it can be adapted for humanoid robots with some modifications.

Nav2 Architecture

The Nav2 stack consists of several key components:

1. Global Planner: Creates a path from start to goal
2. Local Planner: Follows the global path while avoiding obstacles
3. Controller: Converts path following commands to robot controls
4. Recovery Behaviors: Handles navigation failures

Nav2 Launch File for Humanoid Robots

<!-- humanoid_nav2.launch.py -->
from launch import LaunchDescription
from launch.actions import DeclareLaunchArgument, SetEnvironmentVariable
from launch.substitutions import LaunchConfiguration
from launch_ros.actions import Node
from nav2_common.launch import RewrittenYaml

def generate_launch_description():
    use_sim_time = LaunchConfiguration('use_sim_time')
    autostart = LaunchConfiguration('autostart')
    params_file = LaunchConfiguration('params_file')

    lifecycle_nodes = ['controller_server',
                       'planner_server',
                       'recoveries_server',
                       'bt_navigator',
                       'waypoint_follower']

    # Map server parameters for humanoid-specific maps
    map_server_params = {
        'yaml_filename': '/path/to/humanoid_map.yaml',
        'frame_id': 'map',
        'topic_name': 'map',
        'use_bag_pose': False
    }

    return LaunchDescription([
        # Declare launch arguments
        DeclareLaunchArgument(
            'use_sim_time', 
            default_value='false',
            description='Use simulation time if true'),
        DeclareLaunchArgument(
            'autostart', 
            default_value='true',
            description='Automatically start lifecycle nodes'),
        DeclareLaunchArgument(
            'params_file', 
            default_value='/path/to/humanoid_nav2_params.yaml',
            description='Full path to the ROS2 parameters file'),

        # Map server
        Node(
            package='nav2_map_server',
            executable='map_server',
            name='map_server',
            parameters=[map_server_params],
            output='screen'),

        # Local costmap
        Node(
            package='nav2_costmap_2d',
            executable='costmap_2d_node',
            name='local_costmap',
            parameters=[params_file],
            output='screen'),

        # Global costmap
        Node(
            package='nav2_costmap_2d',
            executable='costmap_2d_node',
            name='global_costmap',
            parameters=[params_file],
            output='screen'),

        # Controller server for humanoid-specific movement
        Node(
            package='nav2_controller',
            executable='controller_server',
            name='controller_server',
            parameters=[params_file],
            output='screen'),

        # Planner server
        Node(
            package='nav2_planner',
            executable='planner_server',
            name='planner_server',
            parameters=[params_file],
            output='screen')
    ])

Humanoid Navigation Considerations

Navigating with a humanoid robot requires special considerations:

Bipedal Locomotion

• Footstep Planning: Instead of continuous paths, humanoid robots need discrete footsteps
• Balance Maintenance: Controllers must maintain balance during movement
• Stability: Walking gaits must be dynamically stable
• Terrain Adaptation: Ability to handle uneven terrain

Configuration Example

# humanoid_nav2_params.yaml
amcl:
  ros__parameters:
    use_sim_time: False
    alpha1: 0.2
    alpha2: 0.2
    alpha3: 0.2
    alpha4: 0.2
    alpha5: 0.2
    base_frame_id: "base_footprint"
    beam_skip_distance: 0.5
    beam_skip_error_threshold: 0.9
    beam_skip_threshold: 0.3
    do_beamskip: false
    global_frame_id: "map"
    lambda_short: 0.1
    laser_likelihood_max_dist: 2.0
    laser_max_range: 100.0
    laser_min_range: -1.0
    laser_model_type: "likelihood_field"
    max_beams: 60
    max_particles: 2000
    min_particles: 500
    odom_frame_id: "odom"
    pf_err: 0.05
    pf_z: 0.99
    recovery_alpha_fast: 0.0
    recovery_alpha_slow: 0.0
    resample_interval: 1
    robot_model_type: "nav2_amcl::DifferentialMotionModel"
    save_pose_rate: 0.5
    sigma_hit: 0.2
    tf_broadcast: true
    transform_timeout: 1.0
    update_min_a: 0.2
    update_min_d: 0.25

controller_server:
  ros__parameters:
    use_sim_time: False
    controller_frequency: 20.0
    min_x_velocity_threshold: 0.001
    min_y_velocity_threshold: 0.5
    min_theta_velocity_threshold: 0.001
    progress_checker_plugin: "progress_checker"
    goal_checker_plugin: "goal_checker"
    controller_plugins: ["FollowPath"]

    # Humanoid-specific controller
    FollowPath:
      plugin: "nav2_mppi_controller::MppiController"
      time_steps: 24
      control_freq: 20.0
      horizon: 1.5
      Q: [2.0, 2.0, 0.8]
      R: [1.0, 1.0, 0.5]
      P: [0.02, 0.02, 0.02]
      collision_penalty: 100.0
      goal_angle_tolerance: 0.15
      goal_check_tolerance: 0.25
      inflation_radius: 0.15
      debug_cost_data_enabled: False
      motion_model: "DiffDrive"
      # For humanoid robots, use appropriate motion model

local_costmap:
  ros__parameters:
    use_sim_time: False
    update_frequency: 5.0
    publish_frequency: 2.0
    global_frame: odom
    robot_base_frame: base_link
    footprint: "[ [0.3, 0.3], [0.3, -0.3], [-0.3, -0.3], [-0.3, 0.3] ]"
    resolution: 0.05
    inflation_radius: 0.55
    plugins: ["voxel_layer", "inflation_layer"]
    voxel_layer:
      plugin: "nav2_costmap_2d::VoxelLayer"
      enabled: True
      voxel_size: 0.05
      max_voxels: 10000
      mark_threshold: 0
      observation_sources: scan
      scan:
        topic: /scan
        max_obstacle_height: 2.0
        clearing: True
        marking: True
        data_type: "LaserScan"
    inflation_layer:
      plugin: "nav2_costmap_2d::InflationLayer"
      cost_scaling_factor: 3.0
      inflation_radius: 0.55
    always_send_full_costmap: True

global_costmap:
  ros__parameters:
    use_sim_time: False
    update_frequency: 1.0
    publish_frequency: 1.0
    global_frame: map
    robot_base_frame: base_link
    footprint: "[ [0.3, 0.3], [0.3, -0.3], [-0.3, -0.3], [-0.3, 0.3] ]"
    resolution: 0.05
    track_unknown_space: true
    plugins: ["static_layer", "obstacle_layer", "inflation_layer"]
    obstacle_layer:
      plugin: "nav2_costmap_2d::VoxelLayer"
      enabled: True
      voxel_size: 0.05
      max_voxels: 10000
      mark_threshold: 0
      observation_sources: scan
      scan:
        topic: /scan
        max_obstacle_height: 2.0
        clearing: True
        marking: True
        data_type: "LaserScan"
    static_layer:
      plugin: "nav2_costmap_2d::StaticLayer"
      map_subscribe_transient_local: True
    inflation_layer:
      plugin: "nav2_costmap_2d::InflationLayer"
      cost_scaling_factor: 3.0
      inflation_radius: 0.55

planner_server:
  ros__parameters:
    use_sim_time: False
    planner_plugins: ["GridBased"]
    GridBased:
      plugin: "nav2_navfn_planner::NavfnPlanner"
      tolerance: 0.5
      use_astar: false
      allow_unknown: true

Integration with Isaac ROS Visual SLAM

To integrate Isaac ROS VSLAM with Nav2:

import rclpy
from rclpy.node import Node
from geometry_msgs.msg import PoseWithCovarianceStamped
from sensor_msgs.msg import Image, CameraInfo
from tf2_ros import TransformBroadcaster
import tf_transformations

class IsaacVSLAMIntegrator(Node):
    def __init__(self):
        super().__init__('isaac_vslam_integrator')
        
        # Subscribers for Isaac ROS VSLAM
        self.pose_sub = self.create_subscription(
            PoseWithCovarianceStamped,
            '/visual_slam/pose_graph/pose',
            self.pose_callback,
            10
        )
        
        # Publishers for Nav2
        self.initial_pose_pub = self.create_publisher(
            PoseWithCovarianceStamped,
            '/initialpose',
            10
        )
        
        # TF broadcaster for VSLAM to Nav2 transform
        self.tf_broadcaster = TransformBroadcaster(self)
        
        self.get_logger().info('Isaac VSLAM Integrator initialized')

    def pose_callback(self, msg):
        # Process the VSLAM pose and potentially send to Nav2
        self.get_logger().info(f'VSLAM pose: x={msg.pose.pose.position.x}, y={msg.pose.pose.position.y}')
        
        # Broadcast transform from map to odom using VSLAM data
        t = msg.pose.pose  # Position and orientation from VSLAM
        
        # Create TF message
        from geometry_msgs.msg import TransformStamped
        tf_msg = TransformStamped()
        
        tf_msg.header.stamp = self.get_clock().now().to_msg()
        tf_msg.header.frame_id = 'map'
        tf_msg.child_frame_id = 'odom'
        
        tf_msg.transform.translation.x = t.position.x
        tf_msg.transform.translation.y = t.position.y
        tf_msg.transform.translation.z = t.position.z
        tf_msg.transform.rotation = t.orientation
        
        self.tf_broadcaster.sendTransform(tf_msg)

def main(args=None):
    rclpy.init(args=args)
    integrator = IsaacVSLAMIntegrator()
    rclpy.spin(integrator)
    integrator.destroy_node()
    rclpy.shutdown()

if __name__ == '__main__':
    main()

Real-world Considerations for Humanoid VSLAM

Lighting Conditions

• Indoor Environments: Usually have consistent lighting
• Windows: Can cause lighting changes that affect tracking
• Artificial Lighting: May create shadows that affect feature detection
• Dynamic Lighting: Moving from bright to dark areas

Motion Artifacts

• Head Movement: Humanoid robots often move their heads, affecting camera perspective
• Body Dynamics: Walking motion can cause camera vibration
• Fast Movements: Rapid head movements can cause motion blur

Troubleshooting VSLAM Issues

Tracking Loss

• Solution: Implement relocalization or use sensor fusion with IMU
• Cause: Insufficient visual features or fast movement

Drift

• Solution: Use loop closure detection and pose graph optimization
• Cause: Accumulated errors in pose estimation

Map Quality

• Solution: Optimize parameters and use appropriate sensors
• Cause: Poor lighting, repetitive textures, or dynamic objects

Summary

Isaac ROS provides powerful GPU-accelerated perception capabilities that are especially valuable for humanoid robots. Visual SLAM offers rich environmental understanding that can be integrated with Nav2 for robust navigation. The combination of Isaac ROS and Nav2 enables humanoid robots to navigate complex human environments safely and efficiently.

Exercises

1. Set up Isaac ROS Visual SLAM in a simulation environment
2. Configure Nav2 for a humanoid robot model
3. Integrate the two systems and test navigation

Next Steps

In the next chapter, we'll explore Nav2 path planning specifically adapted for bipedal robots, addressing the unique challenges of humanoid locomotion.