From Pixels to Production: Texas Instruments and NVIDIA Partner to Harden Humanoid Safety

SAN JOSE — As the race to develop general-purpose humanoids shifts from digital "dreams" to factory floors, the focus is increasingly landing on the hardware that keeps these machines from crashing into their surroundings. Ahead of its GTC 2026 conference, NVIDIA has announced a deep technical collaboration with Texas Instruments (TI) aimed at hardening the safety and perception stacks of next-generation physical AI.

The partnership centers on the integration of TI’s mmWave radar technology with the NVIDIA Jetson Thor robotics computer and the NVIDIA Holoscan framework. By combining high-frequency sensing with massive AI compute, the companies aim to provide a "production-ready" foundation for humanoids to operate safely alongside humans in unpredictable environments.

Bridging the Perception Gap

While camera-based systems have seen massive improvements through models like DreamZero, they remain vulnerable to common environmental hazards such as blinding glare, thick dust, or transparent glass doors. TI’s IWR6243 mmWave radar sensors address these "edge cases" by providing 3D perception that is independent of lighting conditions.

According to the announcement, the sensors connect via Ethernet to NVIDIA Jetson Thor using the Holoscan Sensor Bridge. This configuration enables low-latency sensor fusion, allowing a robot to cross-reference visual data with radar pings in real time. This redundancy is critical for navigating complex settings like hospitals or retail spaces where glass partitions and reflective surfaces often "ghost" traditional computer vision systems.

From Simulation to the Fab

This collaboration represents the physical counterpart to NVIDIA’s recent software breakthroughs. While tools like DreamDojo allow robots to learn intuitive physics in a 44,000-hour "dream" state, TI is providing the deterministic hardware required to translate those learned behaviors into the real world.

For Texas Instruments, the deal further cements its role as a key architect in the humanoid supply chain. The company recently established a strategic partnership with UBTECH, which saw TI’s components integrated into the Walker S2 humanoid. That deal was reciprocal, with TI deploying UBTECH machines into its own semiconductor manufacturing lines to handle sensitive material logistics.

Scaling Physical AI

The integration with Jetson Thor—NVIDIA’s specialized SoC for humanoids—signals a move toward standardized, functional safety-capable platforms. Deepu Talla, NVIDIA’s VP of robotics, noted that synchronizing complex AI models with real-time motor controls requires a "massive leap" in processing power.

By offloading environmental sensing to specialized radar and power management subsystems from TI, NVIDIA can dedicate more of its Blackwell-based architecture to high-level reasoning and whole-body control systems like SONIC.

The joint solution will be demonstrated live at NVIDIA GTC, March 16–19, featuring a sensor fusion processing chain developed in partnership with D3 Embedded. As the industry moves toward what researchers call the "GPT-2 moment" for robotics, the TI-NVIDIA tie-up suggests that the most reliable path to "Physical Turing Test" levels of grace isn't just better code, but a more robust sensory nervous system.