Drone autonomy: navigation, BVLOS, detect and avoid, delivery and inspection

A drone that can hold a hover is a flying platform; a drone that can plan a route, fly kilometers beyond its operator's sight, notice an approaching helicopter and step aside, deliver a parcel to a backyard or inspect a power line without help is an autonomous system. This lesson covers the autonomy layer above the flight controller of Lesson 6: mission planning and navigation, operation beyond visual line of sight, detect-and-avoid, and the two largest commercial applications, delivery and infrastructure inspection. Drone regulation is treated in depth in PHOT-350; here we outline only the parts that shape how autonomy is designed, and we note where rules are still changing.

Missions and navigation

Most autonomous drone missions are planned in advance as a sequence of waypoints with altitudes, speeds and actions, such as take a photo, hover or drop a payload, uploaded to the autopilot by a ground control station. Survey missions use lawnmower patterns with overlap chosen for photogrammetry; corridor missions follow linear assets such as pipelines and power lines. During flight, the autopilot navigates with GNSS and its inertial estimator, while the ground station monitors telemetry. Higher autonomy adds on-board path planning around obstacles detected by cameras or lidar, dynamic replanning when a no-fly area is activated, and contingency planning that keeps a reachable landing site available at every point of the route.

GNSS dependence is a weakness. Inspection close to steel structures, flight under bridges and indoor or underground work, such as mine inspection, deny satellite signals and can disturb the magnetometer. Visual-inertial odometry and lidar SLAM, from ROB-340, let drones navigate without GNSS. Jamming and spoofing are also growing concerns, and robust systems cross-check GNSS against inertial and visual estimates.

# Uploading and flying a simple survey mission with the MAVSDK Python library (PX4 autopilot)
import asyncio
from mavsdk import System
from mavsdk.mission import MissionItem, MissionPlan

async def run():
    drone = System()
    await drone.connect(system_address="udp://:14540")      # simulator or telemetry link
    async for state in drone.core.connection_state():
        if state.is_connected:
            break
    items = []
    for lat, lon in [(45.4215, -75.6972), (45.4220, -75.6972),
                     (45.4220, -75.6962), (45.4215, -75.6962)]:
        items.append(MissionItem(lat, lon, 80, 10, True, float('nan'), float('nan'),
                                 MissionItem.CameraAction.TAKE_PHOTO, float('nan'),
                                 float('nan'), float('nan'), float('nan'), float('nan'),
                                 MissionItem.VehicleAction.NONE))
    await drone.mission.set_return_to_launch_after_mission(True)
    await drone.mission.upload_mission(MissionPlan(items))
    await drone.action.arm()
    await drone.mission.start_mission()

asyncio.run(run())

Mission listings like this one are typically tested first in software-in-the-loop simulation, where the autopilot firmware runs on a desktop computer against a simulated vehicle, before any flight. The same mission file can then be flown on hardware, with the ground station watching for failsafe events such as low battery or lost link, which trigger the return-to-launch behavior configured in the autopilot parameters.

Beyond visual line of sight

Under the basic rules in both countries, the pilot must keep the drone within visual line of sight, VLOS, unaided by binoculars. Most economically valuable missions, such as delivery, long linear inspections and large-area agriculture, require flying beyond visual line of sight, BVLOS. Once the pilot cannot see the drone or the sky around it, two functions humans normally perform must be replaced: keeping the drone clear of other aircraft, and keeping it clear of people and obstacles on the ground. Regulators therefore tie BVLOS permissions to the risk of each operation, considering airspace class, population density beneath the route, the drone's mass and reliability, and the detect-and-avoid capability.

Canada's 2025 amendments allow lower-risk BVLOS operations below 122 m in uncontrolled airspace, away from aerodromes and outside populated areas, for pilots holding a Level 1 Complex certificate whose organizations hold an operator certificate; anything more complex needs a Special Flight Operations Certificate. In the United States, the Federal Aviation Administration proposed Part 108 in August 2025 to normalize BVLOS operations under operating permits for lower-risk work and operating certificates for larger or more complex operations, with right-of-way rules relative to crewed aircraft broadcasting their position. The final rule was sent for federal review in July 2026 and, at the time of writing, has not been published and is not in force; until it is, US BVLOS operations rely on waivers and exemptions. Because these details change, always check the current rules and PHOT-350 before planning an operation.

Detect and avoid

Detect-and-avoid, DAA, replaces the pilot's ability to see and avoid other aircraft. It has three parts: surveillance, to detect traffic; threat evaluation, to decide whether a conflict will occur; and maneuver, to resolve it while respecting right-of-way. Surveillance can be cooperative, using signals that aircraft broadcast, such as ADS-B, or non-cooperative, using radar, cameras or acoustic sensors to detect aircraft that broadcast nothing, such as many small private aircraft, gliders and balloons. Ground-based radar networks can cover a whole corridor, while airborne sensors travel with the drone. The required detection range follows from closing speed and the time needed to decide and maneuver.

# Required detect-and-avoid range for different intruders
import numpy as np
drone_v = 15.0                                   # drone ground speed m/s
t_budget = 4.0 + 2.0 + 10.0                       # confirm track, decide, manoeuvre (s)
d_sep = 600.0                                     # desired miss distance (m), roughly 2000 ft
intruders = {"helicopter": 60.0, "light plane": 55.0, "glider": 30.0, "other drone": 15.0}
for name, v in intruders.items():
    v_close = drone_v + v                         # head-on worst case
    R = v_close * t_budget + d_sep
    print(f"{name:12s}: closing {v_close:5.1f} m/s, detect at {R / 1000:4.2f} km")

helicopter  : closing  75.0 m/s, detect at 1.80 km
light plane : closing  70.0 m/s, detect at 1.72 km
glider      : closing  45.0 m/s, detect at 1.32 km
other drone : closing  30.0 m/s, detect at 1.08 km

Low-altitude airspace is also becoming organized. UAS traffic management, UTM, systems let operators share flight intents, receive notices of restrictions and deconflict routes strategically before take-off, and Remote ID broadcasts each drone's identity and position so authorities and others can see it. Strategic deconfliction reduces the burden on tactical DAA but does not replace it, because crewed aircraft may not participate.

Ground risk

The other half of BVLOS safety is ground risk: the chance that a failure injures someone below. It depends on population density under the route, the drone's mass and impact energy, and its failure rate. Mitigations include routing over sparsely populated areas, parachutes that reduce impact energy, flight termination systems, and reliability engineering with redundant propulsion, power and navigation. Risk assessment methods such as the Specific Operations Risk Assessment, SORA, developed by the international JARUS group and used in many jurisdictions including Canada's risk framework, combine ground and air risk classes into a required level of robustness.

Delivery

Drone delivery has moved from trials to routine service in some places. Operators fly small parcels, meals, prescriptions and medical supplies, often from a central hub or a store, with drones that lower packages on a tether or drop them by parachute so the aircraft stays high above people. Fixed-wing and hybrid vertical-take-off designs extend range for medical delivery in remote areas, and remote and northern communities in Canada are a natural market, because roads may be seasonal and distances large. The challenges are weather (wind, rain, icing and cold), noise acceptance, delivery site safety, airspace integration, and economics: a delivery drone must complete many trips per day with few people supervising many aircraft, which drives the push toward one-to-many operations and regulatory approval for them.

Inspection

Inspection is the other big application: power lines, wind turbines, bridges, pipelines, rail, telecom towers, solar farms, roofs and mine stopes. Drones replace rope access, helicopters and scaffolding, reducing risk to workers. Autonomy helps in several ways: flying precise repeatable paths so images from successive inspections can be compared, holding a fixed standoff from structures using lidar or stereo cameras, navigating inside GNSS-denied spaces such as boiler interiors or culverts, and using on-board or cloud machine learning to flag defects such as corrosion, cracks or damaged insulators. Docking stations, drone-in-a-box systems that charge and shelter the aircraft, allow scheduled remote inspection without a crew on site, which is an important use case for BVLOS approvals. Results feed asset management systems, and human experts still review flagged findings.