In the fast-moving world of humanoid robotics, much of the attention remains focused on high-profile prototypes and demonstration videos.
Yet behind the scenes, a quieter but equally critical layer of innovation is taking place – one rooted in semiconductors, sensing, control systems, and the underlying architectures that make physical machines viable at scale.
Texas Instruments is one of the companies operating at this foundational level. Established in 1930 and long associated with American industrial and technological development, TI has evolved from its early days in oil exploration electronics into a global supplier of analog and embedded processing chips.
These days, Texas Instruments designs and manufactures analog and embedded processing chips that control, sense, and manage power in electronic systems across industries including industrial automation, automotive, and consumer devices.
While it no longer commands the same public attention as some newer AI-focused firms, the company remains deeply embedded in modern electronics supply chains – including those increasingly shaping robotics and automation.
As humanoid robots move from research labs toward early commercial pilots, TI is positioning itself as an enabler of what is often described as “physical AI” – systems that must sense, decide, and act reliably in real-world environments.
That shift places new demands on hardware, from deterministic real-time control and high-bandwidth sensing to power efficiency and system-level integration.
In this Q&A, German Aguirre, systems manager for robotics at Texas Instruments, outlines the technical gaps that still separate today’s humanoid prototypes from scalable, production-ready systems.
His responses highlight recurring themes across the industry: the challenge of achieving reliability at scale, the growing importance of end-to-end functional safety, and the difficulty of delivering dexterous, human-like manipulation.
Aguirre also points to the increasing role of sensor fusion – combining vision with technologies such as radar – and the need for tightly integrated hardware architectures capable of supporting real-time decision-making.
Together, these elements form the less visible but essential infrastructure that may ultimately determine how quickly humanoid robots transition from concept to widespread deployment.
Interview with German Aguirre, systems manager for robotics, Texas Instruments
Robotics & AutomationNews: There’s a lot of momentum around humanoid robots, but relatively little real-world deployment. From your perspective, what is the single biggest gap between today’s prototypes and commercially viable systems?
German Aguirre: One of the biggest design challenges is achieving reliability at scale.
Today’s humanoids can demonstrate impressive capabilities in controlled environments, but commercial deployment requires repeatable performance across millions of cycles, varying environments and edge cases.
This gap shows up in three key areas. The first is robust perception, which includes handling lighting, occlusion and dynamic environments.
The second is deterministic real-time control, meaning low-latency, synchronized actuation across many axes. Lastly, system-level power and thermal efficiency, which is essential for humanoids to operate all day.
Addressing this challenge requires tighter integration across sensing, processing and actuation.
R&AN: You emphasize functional safety across the entire signal chain. How different is safety engineering for humanoids compared to traditional industrial robots or automated vehicles?
GA: Safety engineering for humanoid robots is highly important and inherently different because these systems operate in unstructured environments alongside humans, rather than in fenced industrial cells.
For example, humanoids do not have a centralized safety controller. Instead, humanoids require safety mechanisms at the joint, sensor and system level.
Additionally, unlike fixed industrial robots, humanoids humanoids continuously adapt their motion, making failure modes harder to define. Safety in humanoids is also increasingly dependent on real-time perception and sensor fusion, not just hardwired interlocks.
All of this drives the need for end-to-end safety across the entire signal chain, including sensing and communication, real-time control and power delivery.
R&AN: Dexterous manipulation is often described as the hardest problem in robotics. From a hardware and control standpoint, what breakthroughs are still needed to achieve reliable, repeatable performance?
GA: The biggest design challenges in dexterous manipulation are in sensing fidelity and control bandwidth.
Achieving reliable, repeatable performance requires:
- High-resolution force/torque and tactile sensing embedded directly into hands and joints.
- Higher bandwidth motor control, including higher PWM frequencies and faster current loops, to enable smooth, responsive interaction.
- Tightly coupled sensing and control loops for true force-position hybrid control.
One of the most persistent hardware challenges in dexterous manipulation is delivering high power density in extremely compact spaces such as robot hands and wrists.
GaN power stages are making meaningful progress here, achieving greater efficiency compared to traditional silicon-based designs. This efficiency means less heat, smaller passive components and more compact motor drivers.
Reliable, dexterous manipulation also demands real-time motor control – the ability to sense, process, and actuate within microseconds.
Real-time microcontrollers, such as TI’s C2000 and Arm Cortex-based devices, can execute complete current control loops in under one microsecond, enabling the precise positioning needed for tasks like picking up small objects or turning a door handle.
The goal is fully deterministic, low-latency control that allows robots to operate safely and fluidly.
Take TI’s TIDA-010979 and TIDA-010992 reference designs as an example, which feature TI’s Sitara and C2000 microcontrollers for joint and hand control respectively.
These MCUs provide high-performance real-time control, enabling humanoids to operate with the precision, stability and response time needed to meet growing performance demands.
R&AN: “Physical AI” is becoming a widely used term. In practical terms, what does it require at the system level – beyond just more powerful compute?
GA: At the system level, requirements for physical AI go well beyond compute power.
It requires:
- Low-latency sensing and actuation loops (real-time, deterministic control).
- Reliable power architecture to support high peak loads and continuous operation.
- High-speed, deterministic communication between distributed nodes such as joints and sensors.
- Edge intelligence tightly coupled with hardware.
In other words, intelligence only matters if the system can sense, decide, and act in real time with high reliability.
TI’s broad portfolio spanning sensing, processing, control and communication technologies provides the foundation needed for physical AI systems to operate safely, predictably and at scale.
R&AN: Sensor fusion is critical for real-world operation. How do technologies like radar complement vision systems in making humanoids reliable in unpredictable environments?
GA: Vision systems such as cameras are an important sensing technology for robotics. However, they are affected by lighting, dust, occlusion and texture.
Radar complements vision by providing robust detection in all environmental conditions, velocity and motion awareness (direct Doppler measurement) and depth information independent of lighting.
In humanoids, this enables more reliable obstacle detection, redundant sensing for safety and improved tracking in dynamic environments. The key is sensor fusion, where radar adds a layer of robustness that vision alone cannot provide.
TI supports this with solutions like our IWR6243 mmWave radar sensor, which collects radar data that – when fused with camera data – can reduce false positives and improve object detection, localization and tracking for physical AI applications.
Together, these technologies create robust perception systems that can operate safely and reliably across the unpredictable scenarios humanoids will encounter in real-world deployments.
R&AN: TI is working with companies like Apptronik and Nvidia. How important are these ecosystem partnerships in accelerating development, and where do you see the biggest bottlenecks today?
GA: Ecosystem partnerships are very important for humanoid development. As one of the most complex system integration challenges in engineering today, these partnerships enable development by aligning compute, sensing and actuation stacks.
By allowing each company to focus on their core strengths, we can accelerate the path to scalable, production-ready humanoid robots while ensuring seamless integration.
That said, one of the biggest bottlenecks is co-optimization. Hardware and software constraints are deeply interdependent and cannot be solved in isolation. For example, reducing latency often increases compute demand, which in turn drives up power consumption and thermal load.
Other bottlenecks include system integration complexity, power and thermal constraints and reliable, real-time communication across distributed architectures.
Ecosystem partnerships help address these challenges quickly and collaboratively, allowing developers to move from virtual development to production-ready, scalable and safety-compliant systems.
R&AN: Looking ahead, what milestones need to be achieved before humanoid robots can move from pilot deployments to scaled, everyday use in industrial or commercial settings?
GA: Three milestones need to be achieved.
- First, humanoids need to achieve reliable, repeatable task execution. Humanoid robots must perform tasks consistently across different environments, not just in demos.
- All-day operation is also critical. This requires advances in power efficiency, thermal design and battery systems to enable continuous use.
- Lastly, cost-effective and scalable architectures. The industry needs standardized, modular designs that can scale across applications.
Once these are in place, we’ll see humanoids transition from prototypes to broader industrial and commercial deployment.
TI is committed to helping the industry achieve these milestones with our system-level solutions. By providing the foundational semiconductors needed, from perceptive sensing and precise motor control to real-time communication and AI technologies, we are helping OEMs bring humanoids from the lab to everyday life.
