isaac_ros_correlated_timestamp_driver

Overview

An autonomous robotics system needs to perceive itself, its environment, and plan its actions to achieve a task, and control those actions, in a closed loop. Perception is provided by a combination of sensors, for example camera, LIDAR, IMU, touch, motor positions, wheel ticks, and others. Each of these sensors operate independently running at different frequencies and many on their own clocks. This is a challenge when more than one sensor is needed to reconstruct an understanding of the environment for planning of actions.

A common practice is to use the system time as the clock on which all sensor data is tracked in a single timeline. When the sensor’s data is received, an interrupt occurs causing the software to record the sensor acquisition time as the current system time. This lacks accuracy as the current system time does not account for time spent transmitting the sensor data to the processor, or kernel delays in servicing the interrupt which records the system time. The acquisition time of the sensor data will be both late, and subject to jitter. For some applications this lack of accuracy in the single timeline for sensor data is a problem for a precise reconstruction of the environment and precise planning at higher action speeds.

The isaac_ros_correlated_timestamp_driver provides functionality for highly accurate acquisition time of sensor data translated to the system time. This is needed for precise reconstruction of the environment such as 3D maps, cost maps, and obstacle avoidance from multiple concurrent sensors. This enables high action speeds as multi-sensor perception has 2x orders of magnitude lower jitter (from 1ms to <10us).

This accuracy is provided by isaac_ros_correlated_timemestamp_driver by leveraging hardware features in Jetson platforms. A hardware feature uses TSC as the sensor acquisition time when an interrupt is received. This eliminates jitter in sensor acquisition time measurements caused by delays in the kernel servicing interrupts. This is used for camera, and IMU data in Nova. isaac_ros_correlated_timemestamp_driver converts TSC measurements to system time. In addition, time synchronization between system time, and Ethernet is maintained with PTP, using phc2sys. Hardware provides time synchronization between TSC and PTP, for highly precise sensor data acquisition times between camera and LIDAR sensors needed for sensor fusion in 3D reconstruction.

Quickstart

Troubleshooting

API

ros2 launch isaac_ros_correlated_timestamp_driver correlated_timestamp_driver.launch.py target_container:=<Target Container> nvpps_dev_file:=<NVPPS Dev Name> use_time_since_epoch:=<Use Time Since Epoch>