Camera Development#

This section describes the camera software solution included in NVIDIA^® Jetson™ Linux.

NVIDIA Jetson offers SIPL as the major camera software path, while maintaining support for Argus (with no new feature additions). SIPL is the primary camera software beginning with Jetson Thor. The following table summarizes how Argus and SIPL differ in scope, architecture, and direction.

Argus vs SIPL#
Theme	Argus	SIPL
Primary purpose	Per-frame control camera framework.	Focused on a unified streaming-oriented camera framework targeting robotics and humanoid with safety in mind.
Camera and transport support	Primarily CSI (MIPI) cameras; also supports GMSL on Tegra.	Supports GMSL/CoE with JetPack 7.2. Support for MIPI planned for future releases.
API model	Request/capture-session model via LibArgus and Tegra Multimedia (MM) API.	Streaming-optimized, callback-based API built for continuous sensor streams (INvSIPLCamera and Tegra Multimedia API).
Architecture and configuration	C++ interface tightly coupled to Tegra Multimedia stack; Argus-specific driver model.	Modular, extensible, JSON/UDDF-driven configuration; SIPL Core/Query separates camera and transport, with unified CameraHAL and protocol-independent pipeline APIs.
Functionality scope	Strong ISP capture features, including mature multi-camera sync (stereo pipeline with 6-sigma sync on L4T).	Broader “camera integration + streaming + control” scope; stereo added in JetPack 7.2 and multi-camera sync being added in phases.
Hardware ISP	Argus supports hardware ISP on L4T.	SIPL supports two hardware ISP on Jetson Thor.
Performance and resource usage	Designed around earlier ISP workflows; has higher CPU/thread/latency overhead for streaming compared to SIPL.	Explicit goal is a streaming API with lower CPU, fewer threads, and reduced latency for high-performance applications.
Safety and unification with Drive OS	Jetson-only legacy framework; not the primary path for safety-critical use going forward.	Already the automotive safety framework; L4T work explicitly targets a unified architecture stack with Drive OS (same SIPL Core/Control/Query).
Roadmap and lifecycle	Planned to move into sustenance on newer platforms; Argus is not the long-term recommended interface (deprecation expected when SIPL parity is achieved).	All new camera features and investments (CoE/HSB, multi-camera sync, IQ tools, multi-modal integration, and so on) are planned on SIPL.
Driver and extensibility model	Argus-specific drivers tied to Tegra V4L2 stack; less modular for user-mode driver experimentation.	SIPL exposes a clean driver model (UDDF/HSL, user-mode sensor drivers), with user-modifiable blocks and easier debugging from user space.
Scalability and multi-modal vision	Centered on camera ISP; limited notion of multi-modal sensor aggregation.	Roadmap includes image aggregation via SIPL (LiDAR, radar, and others), SIPL RAW for CUDA/Cloud ISP, plus multi-camera sync and stability for CoE and GMSL.