Detect and avoid and the sensor fusion problem standing between autonomous aircraft and the open sky

Detect and avoid (DAA) is the single biggest technical barrier between current autonomous aviation efforts and routine operations in shared airspace. No single sensor can replicate what a human pilot’s eyes and brain do together, so the industry is pursuing sensor fusion — combining radar, electro-optical cameras, ADS-B, and other inputs into one coherent traffic picture. The technology is advancing rapidly, but certification challenges mean full autonomous operations in unsegregated airspace are still seven to ten years away.

Why Is Detect and Avoid So Hard to Solve?

Every certificated aircraft carrying passengers operates under a foundational principle codified in FAR 91.113: see and avoid. The pilot in command looks outside, identifies traffic, and maneuvers to avoid it. Your brain performs pattern recognition, depth perception, threat assessment, and motor response simultaneously, in roughly half a second.

Remove the pilot from the cockpit — or from the equation entirely — and you need a sensor suite that replicates all of that in every condition: clear skies, haze, rain, night, sun glare, against terrain backgrounds, and in congested airspace.

The bar is not equivalence. The FAA will not certify a DAA system that merely matches average human performance. The system must demonstrably exceed it, because the regulatory framework was built around the assumption that a thinking human is always in the loop.

What Sensors Make Up the DAA Stack?

There is no single technology that solves detect and avoid. Each sensor type brings distinct strengths and critical weaknesses.

Radar (typically X-band or Ku-band) detects objects at range regardless of lighting or weather. It provides range and closure rate data. But small airborne radars have limited power and antenna size, producing poor angular resolution. A radar may confirm something is three miles out without pinpointing its exact position in the sky. Low radar cross-section targets — paragliders, balloons, small drones — can disappear into the noise floor.

Electro-optical and infrared cameras deliver excellent angular resolution and can distinguish a target against a cluttered background when image processing algorithms perform well. However, cameras provide no native range information. A dot on the sensor could be a 747 at fifteen miles or a Cessna 150 at two. Range must be inferred from angular rate, which takes time and eats into the reaction window. Performance also degrades in rain, fog, and direct sunlight.

Acoustic sensors use microphone arrays to detect the sound signatures of approaching aircraft. Detection is omnidirectional, but range is very limited, speed-of-sound latency is significant, and wind noise at flight speeds makes airborne use impractical. This technology is more relevant for ground-based vertiport operations than in-flight DAA.

ADS-B (Automatic Dependent Surveillance-Broadcast) provides clean, precise data — position, altitude, velocity, and identification — from every equipped aircraft. No ambiguity, no image processing required. But ADS-B is cooperative only. Gliders, balloons, ultralights, older aircraft operating under exceptions, parachutists, and many drones do not broadcast on 1090 MHz. The FAA has been explicit: a DAA system that only sees cooperative traffic cannot be certified.

How Does Sensor Fusion Actually Work?

Sensor fusion takes the complementary strengths of each sensor and merges them into a single integrated traffic picture. Radar provides all-weather range detection. Cameras deliver precise angular tracking and target classification. ADS-B covers cooperative traffic with high confidence. Inertial data, GPS, and potentially lidar handle close-in operations.

A fusion algorithm weights each sensor’s inputs based on confidence levels and environmental conditions. In clear weather, cameras carry more weight for angular precision. In IMC, radar dominance increases. ADS-B data, when available, anchors the position solution. The output is a unified threat picture the autonomy software can act on.

The critical challenge is that these systems all interact. A system tuned too sensitively generates false alarms that erode operator trust. One tuned too conservatively fails to maneuver in time. Finding the balance is where the hardest engineering lives.

Who Is Leading DAA Development?

The most mature programs trace back to Department of Defense and NASA funding. The Air Force Research Laboratory’s Skyborg program and its successors have driven autonomous wingman concepts requiring robust DAA. NASA’s Advanced Air Mobility campaign has been flight-testing DAA systems with multiple commercial partners.

Iris Automation (San Francisco) has built Casia, a computer-vision-based DAA system trained on millions of flight encounters using machine learning. Their system processes camera feeds in real time to detect both cooperative and noncooperative traffic. Their focus on the hardest edge cases — a white aircraft against white clouds, a dark helicopter against dark terrain — directly addresses scenarios that challenge even experienced human pilots.

Honeywell is developing the RDR-84K, a compact Ku-band airborne radar designed specifically for unmanned aircraft systems. It has been tested on the MQ-9B SkyGuardian and other platforms, targeting detection ranges that give autonomy software adequate maneuvering time.

The ACAS X-U system represents the next generation of collision avoidance for unmanned aircraft. Unlike the deterministic logic in current TCAS, ACAS X-U uses probabilistic decision-making — modeling uncertainty in sensor measurements, predicting traffic positions, and issuing resolution advisories based on collision probability. It is designed specifically to work with the noisy, uncertain data from multi-sensor fusion rather than the clean transponder replies traditional TCAS depends on.

What Does the Certification Path Look Like?

Demonstrating equivalent level of safety requires far more than a few thousand flight hours. The FAA needs to see:

• Millions of encounter hours of statistical data
• Hardware qualification for every sensor
• Software assurance at the highest Design Assurance Levels
• Proof that fusion algorithms contain no failure modes creating blind spots
• Graceful degradation — adequate protection even with one sensor failed

The Beyond Visual Line of Sight Aviation Rulemaking Committee (BVLOS ARC) published framework recommendations in 2022, but translating those into regulations with specific performance standards remains in early stages.

When Will DAA Technology Be Ready?

For small drones at low altitudes in defined corridors, practical DAA enabling routine BVLOS operations is two to three years away. The technology is close, the risk environment is manageable, and companies like Zipline, Wing, and Amazon Prime Air are driving intense economic pressure for approval.

For larger autonomous aircraft — autonomous cargo planes or passenger-carrying eVTOL vehicles operating without an onboard pilot — routine operations in unsegregated airspace are realistically seven to ten years out. Initial deployments will likely be confined to controlled corridors with known traffic density, expanding progressively as confidence builds.

How Will This Technology Benefit Piloted Aircraft?

The sensor fusion work being done for autonomous DAA will directly improve safety for piloted aircraft. The same technology can be integrated into cockpit panels as a next-generation traffic awareness system — combining ADS-B data with radar returns and camera-based visual detection into a comprehensive traffic picture superior to what human eyes alone provide.

This capability does not require autonomy. It simply gives the pilot in command a better tool, and it is already in development.

Key Takeaways

• No single sensor solves detect and avoid. Radar, cameras, ADS-B, and other sensors must be fused into one coherent traffic picture, with each weighted by confidence and conditions.
• Noncooperative traffic is the core problem. The FAA will not certify a system that only detects aircraft broadcasting ADS-B. Gliders, balloons, ultralights, and unequipped aircraft must be detectable.
• Certification is harder than the technology. Demonstrating equivalent-or-better safety than human pilots requires millions of encounter hours, highest-level software assurance, and graceful degradation proof.
• Small drone BVLOS operations are nearest-term (two to three years), while large autonomous aircraft in shared airspace are seven to ten years away.
• Piloted aircraft will benefit too. DAA sensor fusion technology will become next-generation cockpit traffic awareness tools for human pilots.