Target Following Demo

This tutorial demonstrates how to use the AgileBot robot for target following tasks.

Target Following Demo

Overview

The target following demo shows how robots can use the RMPflow motion policy to follow moving targets. This demo uses Isaac Sim's motion generation framework.

Prerequisites

• Completed Isaac Sim Environment Configuration
• Completed Visualizing AgileBot Robot Models

Running the Demo

Using the Demo Script Provided by the Project

cd agilebot_isaac_sim
python agilebot_integration/demos/follow_target.py

Demo Description

RMPflow Motion Policy

RMPflow (Robot Motion Policy Flow) is a framework for robot motion planning that generates smooth, safe trajectories by combining multiple motion policies.

Target Following Algorithm

1. Target Detection: Detect the position of the target object
2. Trajectory Planning: Use RMPflow to compute trajectories from current position to target position
3. Motion Execution: Control robot joints to follow the planned trajectory
4. Obstacle Avoidance: Avoid obstacles during movement

Configuration Parameters

Motion Policy Configuration

Motion policy configuration files are located in the project source directory agilebot_integration/motion_policy_configs/Agilebot/gbt_c7a/ .

According to the Environment Configuration Guide, you need to copy the configuration files to the IsaacSim installation directory:

cp -r agilebot_integration/motion_policy_configs/Agilebot \
  ~/isaacsim/exts/isaacsim.robot_motion.motion_generation/motion_policy_configs/

After copying, the full configuration file path is:

~/isaacsim/exts/isaacsim.robot_motion.motion_generation/motion_policy_configs/Agilebot/gbt_c7a/

Main configuration parameters (reference RMPflow Official Tuning Guide):

Parameter	Description	Default Value
c-space_target_rmp/metric_scalar	C-space target RMP weight	1-100
c-space_target_rmp/robust_position_term_thresh	Robust position term threshold	Adjust based on joint count
target_rmp/metric_scalar	Target RMP weight	Adjust based on task
target_rmp/min_metric_alpha	Minimum metric alpha	0 or non-zero
target_rmp/metric_alpha_length_scale	Metric alpha length scale	100000
target_rmp/proximity_metric_boost_length_scalar	Proximity metric boost length scalar	1
target_rmp/max_metric_scalar	Maximum metric scalar	Large value
target_rmp/accel_p_gain	Acceleration P gain	Adjust based on task
target_rmp/accel_d_gain	Acceleration D gain	Adjust based on task
target_rmp/accel_norm_eps	Acceleration normalization epsilon	Adjust based on task
collision_rmp/metric_scalar	Collision avoidance RMP weight	Comparable to target_rmp
damping_rmp/inertia	Damping RMP inertia	0

Tuning Suggestions

1. C-space Target RMP: Set metric_scalar in the range 1-100, which sets the global scale of all RMPs
2. Target RMP: Set a large max_metric_scalar to make it dominant; C-space target will operate in the nullspace of the target RMP
3. Collision Avoidance RMP: Set weight comparable to target_rmp/max_metric_scalar
4. Directional Term: Set min_metric_alpha to a non-zero value and adjust metric_alpha_length_scale for good behavior

FAQ

Q: Robot cannot reach target position

A: Check if the target position is within the robot's workspace. If the target position is outside the workspace, the robot cannot reach it. Adjust target_tolerance parameter to suit task requirements.

Q: Motion trajectory is not smooth

A: Adjust RMPflow configuration parameters, especially:

• Check the gain values of target_rmp/accel_p_gain and target_rmp/accel_d_gain
• Ensure metric_scalar values are within a reasonable range

Q: Robot collides with obstacles

A: Increase collision_rmp/metric_scalar parameter value to ensure the collision avoidance RMP has sufficient weight to avoid obstacles.

Q: How to tune from scratch

A: Reference the official tuning process:

1. Turn off all RMPs (set metric_scalar to 0)
2. Set all inertia terms to 0
3. Re-enable RMPs one by one: c-space_target_rmp → target_rmp → collision_rmp → axis_target_rmp

Next Steps

• Pick and Place Demo
• Stacking Demo
• Kinematics Verification