News
NVIDIA Holoscan 5 introduces low-latency robotic control, enabling gravity-compensation loops for applications like co-bots and object manipulation.
NVIDIA has unveiled Holoscan 5, a platform designed to streamline low-latency robotic control and high-throughput perception. A key feature demonstrated is its ability to enable gravity-compensation loops, which are crucial for applications such as co-bots, object manipulation, and robots learning by demonstration. This capability was showcased using the Seeed Studio reBot Arm B601-DM, a platform equipped with quasi-direct-drive actuators for high torque bandwidth.
Gravity Compensation with Holoscan 5
Gravity-compensation loops allow a robot arm to maintain its position while feeling weightless to the user. This is achieved by continuously calculating the inverse dynamics of the system, incorporating precise models of the robot arm’s links, dimensions, masses, and moments of inertia. The Holoscan 5 SDK simplifies this process by integrating custom operators into a single execution graph, reducing latency and jitter.
The demonstration utilized the Pinocchio library as the inverse-dynamics solver, which calculates joint torques based on the robot’s state at 400 Hz. Commands were sent to the arm’s motors using the motorbridge library, ensuring smooth and responsive operation. Holoscan partitions grouped operators on a single thread, avoiding cross-thread handoffs and minimizing runtime overhead.
Performance and Profiling
Holoscan 5’s built-in Nsight Systems profiling tools provided detailed insights into latency and jitter. By pinning the control partition to a dedicated CPU core, latency was reduced significantly—by approximately 5x for C++ implementations and 9.6x for Python implementations. This optimization ensures the stability and responsiveness required for torque-controlled applications.
Accessible Development Platform
The Seeed Studio reBot Arm B601-DM served as the hardware platform for the tutorial. Unlike traditional low-cost robot arms that rely on hobby servos for positional control, the reBot Arm’s quasi-direct-drive actuators enable precise torque control, making it suitable for haptic and collaborative robotics projects. The tutorial provided a step-by-step guide for developers to implement gravity-compensation loops using Holoscan’s Python API, with a complete C++ implementation also available.
Holoscan 5’s lightweight programming model allows developers to integrate existing algorithms without extensive rewrites. Its transparent runtime and timing contracts enable rapid iteration and debugging, making it an accessible platform for modern robotics control.
News
NVIDIA Holoscan 5 introduces low-latency robotic control, enabling gravity-compensation loops for applications like co-bots and object manipulation.
NVIDIA has unveiled Holoscan 5, a platform designed to streamline low-latency robotic control and high-throughput perception. A key feature demonstrated is its ability to enable gravity-compensation loops, which are crucial for applications such as co-bots, object manipulation, and robots learning by demonstration. This capability was showcased using the Seeed Studio reBot Arm B601-DM, a platform equipped with quasi-direct-drive actuators for high torque bandwidth.
Gravity Compensation with Holoscan 5
Gravity-compensation loops allow a robot arm to maintain its position while feeling weightless to the user. This is achieved by continuously calculating the inverse dynamics of the system, incorporating precise models of the robot arm’s links, dimensions, masses, and moments of inertia. The Holoscan 5 SDK simplifies this process by integrating custom operators into a single execution graph, reducing latency and jitter.
The demonstration utilized the Pinocchio library as the inverse-dynamics solver, which calculates joint torques based on the robot’s state at 400 Hz. Commands were sent to the arm’s motors using the motorbridge library, ensuring smooth and responsive operation. Holoscan partitions grouped operators on a single thread, avoiding cross-thread handoffs and minimizing runtime overhead.
Performance and Profiling
Holoscan 5’s built-in Nsight Systems profiling tools provided detailed insights into latency and jitter. By pinning the control partition to a dedicated CPU core, latency was reduced significantly—by approximately 5x for C++ implementations and 9.6x for Python implementations. This optimization ensures the stability and responsiveness required for torque-controlled applications.
Accessible Development Platform
The Seeed Studio reBot Arm B601-DM served as the hardware platform for the tutorial. Unlike traditional low-cost robot arms that rely on hobby servos for positional control, the reBot Arm’s quasi-direct-drive actuators enable precise torque control, making it suitable for haptic and collaborative robotics projects. The tutorial provided a step-by-step guide for developers to implement gravity-compensation loops using Holoscan’s Python API, with a complete C++ implementation also available.
Holoscan 5’s lightweight programming model allows developers to integrate existing algorithms without extensive rewrites. Its transparent runtime and timing contracts enable rapid iteration and debugging, making it an accessible platform for modern robotics control.
