1 Introduction

The Texas Instruments (TI) TDAx family of devices is well-suited for the companion computer function of drone systems to offload edge computing, artificial intelligence (AI), and computer vision tasks. This scalable family of SoC devices has been designed for automotive vision, sensor fusion, and autonomous driving in automotive driver assist systems (ADAS). While this paper focuses on airborne systems, the highlighted TDAx device capabilities can also benefit maritime applications and unmanned ground vehicles.

Modern autonomous unmanned aircraft systems (UASs) rely on a two-processor architecture that separates flight-critical functions from higher-level intelligence as shown in Figure 1-1.

The flight controller - typically a dedicated microcontroller running real-time firmware - manages low-latency stabilization, attitude estimation, and motor control. The companion computer sits alongside it as a more powerful co-processor responsible for computationally intensive tasks that exceed the capabilities of the flight controller: sensor fusion across cameras and lidar, deep learning inference for object detection and avoidance, simultaneous localization and mapping (SLAM), mission planning, and high-bandwidth telemetry processing. The companion computer communicates with the flight controller over a serial link, translating high-level navigation decisions into velocity or position setpoint commands. As drone autonomy requirements expand from simple waypoint following to real-time scene understanding and adaptive replanning, the companion computer has become the determining factor in overall system intelligence - driving demand for processors that combine high computational throughput with the strict size, weight, power, and environmental constraints inherent to airborne platforms.

Regulatory frameworks governing commercial drone operations are simultaneously expanding the scope of permissible missions and raising the bar for system capability. FAA BVLOS waivers [7], EASA's U-Space framework [8], and Remote ID mandates [9] require drones to perform real-time detect-and-avoid, encrypted telemetry broadcast, and integration with unmanned traffic management (UTM) systems [10] - workloads that compound the AI and sensor fusion demands already placed on the companion computer. These requirements favor processors with headroom to spare: a companion computer running near its compute ceiling to process sensor data has no capacity to add compliance workloads without a platform redesign. The TDAx family's heterogeneous architecture - with dedicated acceleration for AI inference, depth processing, and general compute running concurrently - is practical for absorbing these additional regulatory workloads without displacing core autonomy functions.

Edge compute requirements similar to those of the drone companion computer have emerged in the automotive industry as car manufacturers rapidly deploy machine learning hardware in ADAS for tasks such as object detection, driver monitoring, and collision avoidance [1]. These ADAS compute platforms are highly constrained in terms of size, weight, power and cost (SWaP+C), and increasingly focused on improving energy efficiency. Ongoing investments driven by consumer demand, government safety regulations, and a competitive race to achieve Level 5 vehicle autonomy have established the market conditions in which TI TDAx processors are widely used to bring edge compute and AI acceleration to production ADAS solutions worldwide. The TDAx devices integrate general-purpose compute clusters and digital signal processing compute clusters for executing traditional signal processing and machine learning inference algorithms, in compact packages that can operate in space-constrained systems while utilizing passive cooling. These ADAS market-driven SWaP+C optimizations can now be leveraged to innovate and optimize UAS systems.

Several processor platforms are available to drone designers for companion computer applications, and the choice of processor has significant implications for system reliability, development cost, and long-term supply assurance. Consumer and embedded computing platforms designed primarily for cloud-edge or robotics applications typically target benign operating environments, and their industrial temperature ratings, passive thermal management capabilities, and supply continuity commitments reflect those priorities. By contrast, the TDAx family was designed from the outset to operate in the demanding thermal, vibration, and humidity environments of the automotive cabin and under-hood - conditions that closely mirror the environmental stresses encountered by airborne platforms. With a junction temperature range of -40℃ to 125℃, passive cooling capability, and AEC-Q100 automotive qualification, TDAx offers a reliability advantage that consumer-grade AI accelerators lack. Unlike consumer silicon, which requires significant derating or active thermal management, TDAx avoids the associated payload and power budget penalties critical to drone performance.