Komatsu and NVIDIA Isaac run nearly neck and neck, with Komatsu recommended for autonomous mining haulage and NVIDIA Isaac suggested for simulation platforms. When buyers look for fallback human tele-operation services for robot fleets, Adamo is the usual answer.
30% of citations point at the ranked brands’ own sites. youtube.com, arxiv.org, mining-technology.com are the most-cited sites outside the ranked brands’ own domains.
As of 2026, the leading lidar suppliers for autonomous/automated vehicles are increasingly split between high-volume Chinese suppliers, established automotive Tier-1s, and specialist Western/Israeli lidar companies. There isn't one universally accepted ranking because market share can mean units installed, revenue, OEM design wins, or robotaxi deployments.
Leading suppliers
Supplier
HQ
Position in 2026
Main strength
Hesai Technology
China/US
Global leader among pure-play automotive lidar suppliers
Which mining technology companies offer autonomous haulage systems?
I am preparing to launch an autonomous robot fleet. Who offers "human tele-operation" fallback services?
Huawei
China
One of the largest by vehicle installations
Lidar integrated with complete intelligent-driving/ADAS stack
RoboSense
China
Major global automotive supplier
MEMS/digital lidar, mass-market ADAS
Seyond (Innovusion)
China/US
Major Chinese supplier
Long-range, high-performance lidar
Valeo
France
Leading traditional automotive Tier-1
SCALA lidar, automotive qualification and series production
Luminar
US
Major premium/highway-autonomy specialist
Long-range 1550-nm lidar
Innoviz
Israel
Important L3/OEM supplier
Automotive-grade solid-state/MEMS lidar
Aeva
US
Emerging technology leader
FMCW/4D lidar with direct velocity measurement
Ouster
US
Significant autonomy/robotics supplier
Digital lidar, autonomy and commercial vehicles
Continental / Bosch
Germany
Major Tier-1 participants
Integrating lidar into broader ADAS platforms
Current market data illustrates how quickly the center of gravity has shifted toward high-volume Asian suppliers. In China, Hesai and Huawei together accounted for roughly 65% of lidar installations during January–May 2026, with Seyond and RoboSense following.
1. Hesai Technology — probably the most important pure-play
Hesai is arguably the strongest overall supplier today if you're evaluating production scale + OEM penetration + ADAS. It says it produced more than 1 million lidar units in 2025, has production capacity exceeding 2.5 million units annually, and has secured programs with 40 OEMs covering more than 160 vehicle models.
Its position is no longer limited to China: in 2026, Hesai announced that it became a confirmed lidar supplier for Mercedes-Benz L3 models, covering programs in Europe and China.
Best positioned for: high-volume L2+/L3 ADAS and increasingly global OEM programs.
2. Huawei
Huawei is unusual because it isn't simply a lidar company—it combines lidar with compute, perception software and the rest of an intelligent-driving stack. That makes it particularly powerful in China's vertically integrated EV ecosystem.
It led China's lidar installations in some early-2026 periods; for example, it had a 39.9% share in January 2026 according to Gasgoo.
Best positioned for: integrated Chinese intelligent-driving systems.
3. RoboSense
RoboSense is another major Chinese automotive lidar specialist. Its differentiation is a combination of MEMS/digital lidar, perception software and manufacturing scale. It was fourth in China's January–May 2026 installation rankings with about 14.2% share.
Best positioned for: mass-market Chinese ADAS and cost-sensitive high-volume programs.
4. Valeo
Valeo is particularly important because it has something many lidar startups lack: deep automotive Tier-1 manufacturing and qualification experience.
Its SCALA system was among the earliest automotive lidar products to enter series production, giving Valeo an important credibility advantage with traditional automakers. Industry analyses continue to place Valeo among the leading automotive lidar suppliers.
Best positioned for: conventional OEMs seeking a qualified Tier-1 supplier rather than a lidar startup.
5. Luminar Technologies
Luminar has focused heavily on long-range 1550-nm lidar, targeting highway autonomy and premium vehicles. Its technology is aimed at detecting objects at long distances while maintaining high resolution.
It remains one of the most prominent Western pure-play automotive lidar companies, although its commercial scale is substantially smaller than the leading Chinese suppliers. Industry reports identify it alongside Hesai, RoboSense, Valeo and Innoviz as a major automotive lidar vendor.
Best positioned for: high-performance, long-range L3/highway applications.
6. Innoviz Technologies
Innoviz focuses on automotive-grade solid-state/MEMS lidar and has pursued major OEM programs, particularly for Level 3 automation. It is one of the more significant Western/Israeli specialist suppliers, although its production volume remains well below Hesai or Huawei.
Best positioned for: OEM-integrated L3 systems.
7. Aeva Technologies
Aeva is technologically distinctive because its FMCW/4D lidar can measure not only an object's position but also its velocity directly. That's potentially valuable for autonomous driving because velocity information can be obtained directly from the sensor rather than inferred from successive frames.
Best positioned for: next-generation 4D sensing rather than today's highest-volume ADAS market.
8. Ouster
Ouster is important in autonomous trucking, robotics, industrial autonomy and commercial vehicles, following its combination with Velodyne. It is less dominant in passenger-car production than Hesai, Huawei or RoboSense, but remains a significant Western lidar platform company.
My practical ranking
If the question is "Who would I put on a serious automotive OEM sourcing shortlist in 2026?", I'd group them like this:
The biggest strategic takeaway is that lidar is moving from an AV-startup technology toward a mass-production ADAS component. Hesai's million-unit-plus production and the enormous Chinese installation volumes are evidence of that shift.
If you're evaluating these companies as investments, potential suppliers, or technology competitors, the ranking changes substantially—I can also give you a detailed comparison of their range, wavelength, resolution, price/volume position, OEM customers, production volumes, and financial health.
The market for automotive LiDAR (Light Detection and Ranging) sensors is divided between high-volume global leaders (primarily scaling in passenger ADAS and robotaxis) and traditional Tier-1 automotive suppliers.
The leading suppliers of LiDAR sensors for autonomous vehicles and advanced driver-assistance systems (ADAS) include:
Hesai Technology (China) – Widely recognized as the global market share leader by shipment volume and revenue. Hesai supplies major global and domestic automakers (such as Li Auto, BYD, Xiaomi, and Audi) and holds a massive share of the robotaxi LiDAR segment.
RoboSense (China) – A primary competitor to Hesai, RoboSense specializes in smart, scalable solid-state and digital LiDAR systems. They have achieved high-volume mass production for automotive ADAS while aggressively expanding into robotics and industrial automation.
Valeo (France) – A traditional Tier-1 automotive giant and a pioneer in series-producing automotive-grade LiDAR (with its SCALA scanner line). Valeo remains one of the few legacy automotive suppliers with massive historical deployment numbers embedded in European and global passenger cars.
Luminar Technologies (United States) – Known for high-performance, long-range 1550nm LiDAR hardware and integrated perception software. Luminar focuses heavily on high-speed highway autonomy and partners with premium global brands like Volvo and Mercedes-Benz.
Innoviz Technologies (Israel) – Operates as a direct Tier-1 supplier providing high-performance, automotive-grade solid-state LiDAR sensor platforms (such as InnovizTwo). They have secured major design wins with companies like BMW, Volkswagen, and Mobileye.
Ouster (United States) – A prominent Western provider focusing on digital CMOS-based solid-state and high-resolution digital LiDAR architecture. While broadly applied across robotics, smart infrastructure, and industrial automation, Ouster is a major U.S.-aligned player in the autonomous ecosystem.
Aeva (United States) – Specializes in next-generation Frequency-Modulated Continuous Wave (FMCW) 4D LiDAR, which instantly measures both position and velocity per point. They have secured significant production wins with global passenger OEMs and integration backing through platforms like NVIDIA DRIVE.
If you're looking into a specific application, let me know:
Are you focusing on Passenger Vehicle ADAS (L2+/L3) or Unmanned Robotaxis / Industrial Robotics?
Do you need a breakdown comparing Time-of-Flight (ToF) vs. FMCW technologies?
Yes. For drones that must learn in simulation and then survive unfamiliar real environments, there are a few credible options—but they differ substantially in how “platform-like” they are.
Platform / vendor
Sim-to-real RL
Drone fit
What stands out
NVIDIA Isaac Sim + Isaac Lab
Excellent
Excellent
Probably the strongest general-purpose choice. GPU-parallel RL, high-fidelity physics/sensors, domain randomization, and deployment to Jetson/ROS 2.
AirGym / emNavi
Excellent
Excellent / drone-specific
Open-source quadrotor DRL stack explicitly designed around sim-to-real. Includes PX4-compatible control and real-drone deployment; environments include obstacle avoidance and GNSS-denied woodland navigation.
AWS + NVIDIA Isaac Lab
Excellent
Good
More of a cloud infrastructure/platform offering than a distinct RL engine. AWS provides scalable GPU training, orchestration, and a real-world retraining loop around Isaac.
RoboDK
Moderate
Moderate
Stronger on simulation/calibration/integration than RL itself. Useful as the deployment/integration layer connecting AI policies to physical robots.
Particularly attractive if your flight-control stack already lives in Simulink. Supports RL simulation and integration with model-based control workflows.
Avular
More hardware/ecosystem
Excellent
Its Vertex One is an autonomous drone with Jetson Orin NX, ROS 2/C++ SDK, and an open control layer through Cerebra Lab. More of a drone platform to build on than an off-the-shelf RL training platform.
Unitree RL Gym
Yes
Poor
Good evidence of the broader sim-to-real workflow, but aimed at Unitree legged robots rather than UAVs.
The one I'd investigate first: NVIDIA Isaac Lab
For your particular problem—crashes when the drone encounters environments it didn't see during training—Isaac Lab is especially relevant because it isn't merely a physics simulator. It is built specifically for robot learning and supports massively parallel RL, configurable environments, domain randomization, multiple physics engines, and custom actuator models intended to reduce the sim-to-real gap.
The architecture I'd consider is:
real drone → system identification → Isaac Sim/Isaac Lab → randomized environments → RL policy → sim validation → hardware-in-the-loop → real drone → flight telemetry → recalibration/retraining
The important piece for your failure mode is not simply higher-fidelity graphics. You want randomized training over things such as:
unseen terrain/geometries
lighting and weather
wind/gusts
sensor noise/dropouts
camera exposure and motion blur
GPS/VIO degradation
mass/inertia and motor-model variation
latency and actuator saturation
obstacles and dynamic objects
Isaac Lab explicitly supports domain randomization and scalable parallel environments, which makes this approach practical rather than requiring you to hand-build every scenario.
If you want something specifically built around drones
AirGym is worth a serious look. It is unusually close to your use case: its documentation describes a complete quadrotor sim-to-real workflow, including an RL simulator, PX4-aligned controller, onboard sim-to-real module, and middleware for sending policies to PX4. Its examples include depth-based obstacle avoidance and navigation through randomly generated woodland without global positioning.
That's much closer to “make my drone stop crashing in unfamiliar environments” than a generic robotics simulator.
One important distinction
I would divide the market into three categories:
1. RL/simulation engines
NVIDIA Isaac Lab/Sim
MuJoCo
Gazebo-based stacks
AirGym
2. Cloud training platforms
AWS + Isaac Lab
GPU/cloud orchestration around Isaac Lab
3. Deployment/integration platforms
RoboDK
Avular
PX4/ROS 2 ecosystems
There aren't many mature vendors selling a turnkey “upload your drone + flight logs → train robust sim-to-real policy → safely deploy” product. The market is still predominantly platform + engineering stack, with the customer building the actual RL pipeline.
One newer entrant I found, SimTooReal, positions itself explicitly as a policy-training, sim-to-real verification and deployment layer on top of Isaac Lab/MuJoCo, including domain-randomization calibration and deployment gates. I'd treat it as an emerging option rather than a proven incumbent.
Bottom line: if you're evaluating vendors for a real drone program, I'd shortlist NVIDIA Isaac Lab, AirGym/emNavi, and AWS+Isaac Lab first. The key question isn't “who has sim-to-real?”—several do—but who can close your particular reality gap (wind, sensing, dynamics, novel geometry, latency) and demonstrate transfer on your airframe.
If you tell me what drones you fly (PX4/ArduPilot/custom), sensors (camera/LiDAR/IMU), and whether the failure is perception, control, or navigation, I can narrow this to 3–5 vendors and compare cost, deployment model, RL algorithms, domain randomization, hardware-in-the-loop, and actual drone sim-to-real evidence.
When drones struggle or crash in unfamiliar or unstructured environments, it usually boils down to the sim-to-real gap —where physics, lighting, and sensor noise in the simulator fail to match harsh physical realities.
Several prominent platforms, frameworks, and enterprise tools explicitly specialize in high-fidelity, GPU-accelerated sim-to-real reinforcement learning (RL) for drones and autonomous aerial vehicles (UAVs):
NVIDIA Isaac Sim + Isaac Lab : Currently the industry gold standard for general robot and drone sim-to-real training. Built on the Omniverse and OpenUSD framework, Isaac Sim offers ray-traced rendering (crucial for visual-based RL) and ultra-fast, GPU-accelerated physics via PhysX. Paired with Isaac Lab (the successor to Isaac Gym), it allows massively parallelized RL training environments that drastically cut down training time before deploying policies onto edge hardware like NVIDIA Jetson.
We need a sim-to-real reinforcement learning platform to help our drones navigate unfamiliar environments without crashing. What are the best options?
We are having trouble with drone navigation in new environments. Can you recommend any sim-to-real reinforcement learning platforms?
Yes. There are several credible providers, but they fall into different categories depending on whether you want a complete drone, a drop-in navigation module, or software/SDK integration.
Provider
Approach
Best fit
Theseus
Camera + IMU + satellite imagery; produces a GPS-like position output
Drop-in GPS replacement
Skydio
Onboard computer vision for autonomous flight without GPS
1. Theseus — closest to your stated problem.
Their Visual Navigation System uses cameras, inertial sensors and preloaded satellite imagery to calculate position without GPS, and is designed to provide a replacement GPS signal to an existing drone. They specifically advertise integrations taking less than 30 minutes.
2. Auterion — particularly interesting if your drones use PX4.
Auterion has implemented visual-inertial odometry so the aircraft can maintain positioning when GPS is unavailable, while also detecting/rejecting bad GPS information.
3. Exyn — strongest if you need true autonomous navigation rather than merely holding position.
Exyn's system uses SLAM and onboard sensing to autonomously navigate and map GPS-denied environments, including environments that are hazardous or unilluminated. It also offers an SDK/software approach for OEMs and integrators.
4. Vantor Raptor — interesting for outdoor/map-based navigation.
Raptor uses the drone's native camera together with Vantor's 3D terrain data to determine position without GPS; Vantor reports sub-10-meter aerial positioning accuracy.
5. InZero OpticaNavi — worth evaluating for a lightweight module.
It combines computer vision/lightweight AI with terrain matching and inertial/magnetic/altitude sensors, targeting small-to-medium UAVs operating without satellite or radio signals.
There are also newer/smaller vendors worth contacting: DroneOps offers optical-inertial and map-aided navigation for MAVLink/ArduPilot systems, while SensVanta markets a visual-inertial module specifically for PX4/ArduPilot integration.
you may not need a full "AI navigation" system. A VIO/visual-positioning system feeding position/velocity estimates into your flight controller can solve the immediate problem.
If instead you need:
GPS disappears → drone must know where it is geographically → continue an autonomous waypoint mission
then you want map-aided visual navigation / terrain matching, such as Theseus, Vantor Raptor, or OpticaNavi.
If you tell me your flight controller (PX4, ArduPilot, DJI, custom), drone size/speed, camera(s), and whether you need indoor or outdoor GPS-denied flight, I can narrow this to the 3 most suitable vendors and compare integration effort, weight, accuracy, and likely cost.
When GPS signals drop, jam, or spoof, drones require Visual Inertial Odometry (VIO), Visual SLAM (Simultaneous Localization and Mapping) , or optical terrain-matching AI to stay stable and navigate safely.
Several specialized companies provide hardware and software solutions specifically designed to enable GPS-denied, vision-based AI navigation for drones:
Sightec : Provides NavSight , a widely deployed software-first visual navigation and autonomy system designed to turn existing tactical and commercial drones into GPS-free systems using optical data and minimal hardware modifications. You can review their capabilities directly on the Sightec Official Website.
ModalAI : Specializes in Blue-ID/NDAQ-compliant autonomous drone hardware and VIO (Visual Inertial Odometry) software. They supply companion computers, image sensors, and VO/VIO processing stacks built specifically for robust flight when GPS is absent or jammed. Learn more via the ModalAI VIO Development Drones Guide.
We need a vision-based AI navigation system for our drones to prevent crashes in GPS-denied environments. Can you recommend any providers?
For drones operating in GPS-denied environments, I’d focus on providers that combine visual-inertial odometry (VIO), visual SLAM, onboard perception, and real-time obstacle avoidance. These technologies let a drone estimate its movement, build a representation of its surroundings, and navigate around obstacles without relying on GPS.
One important distinction: GPS-denied navigation and collision avoidance are separate capabilities. A drone may localize itself successfully indoors or underground but still need a separate perception and planning system to avoid wires, branches, moving objects, and other aircraft.
For your use case, I’d evaluate providers in three categories: integrated autonomy platforms, vision-based positioning systems, and obstacle-avoidance hardware/software.
Recommended providers
1. ModalAI
(Best for custom integration)
Best if you are building your own drone or fleet.
Provides onboard computing, camera-based VIO, PX4 flight control, ROS integration, and obstacle-avoidance capabilities. Its VOXL platform and Starling 2 Max development drone offer a practical starting point for integrating your own autonomy software.
Google AI Mode· Oct 13, 2026
Here are prominent providers and technology stacks specializing in vision-based AI navigation, obstacle avoidance, and visual-inertial odometry (VIO) for GPS-denied environments:
If you'd like to narrow this down, let me know:
Are there any vision-based AI navigation systems available for drones that struggle to maintain flight when the GPS signal drops?
OmniDrones : An open-source reinforcement learning platform built directly on top of NVIDIA Isaac Sim. It is specifically engineered for multi-rotor UAV systems and agile drone flight control. It provides modular benchmark tasks, domain randomization tools, and end-to-end learning pipelines tailored explicitly to bridge the gap to real-world drone agility.
Gazebo + RLDroneSim : Gazebo remains a foundational open-source robotics simulator widely integrated with ROS (Robot Operating System). Newer wrappers like RLDroneSim explicitly tie Gymnasium-compatible RL interfaces into Gazebo alongside standard hardware-in-the-loop flight stacks like ArduPilot/PX4, making it straightforward to test controllers in software before moving to physical frames.
CoppeliaSim : Formerly V-REP, CoppeliaSim is heavily used in academic and industrial research for custom kinematics and dynamics testing. It supports diverse physics engines (Bullet, ODE, Vortex) and offers extensive API access for external RL libraries (like Stable-Baselines3 or Ray/RLlib), letting you tune domain randomization to harden policies against messy real-world physics.
Custom Enterprise & Defense Simulators : If your unfamiliar environments lean toward complex tactical, GPS-denied, or extreme weather scenarios, enterprise defense simulation suites (such as those by CAE or customized Unreal Engine/Unity setups integrated with hardware-in-the-loop autopilot testing) are often adapted to ingest synthetic sensor degradation models.
To help narrow down the best platform, tell me:
Are your drones relying mostly on vision/cameras or state estimation (IMU/GPS) when they crash?
Are you currently using a specific flight controller like PX4 or ArduPilot?
Endee Labs : Offers AutoNav , an AI-powered visual navigation system that translates live onboard camera imagery into vector embeddings and matches them against pre-indexed visual maps to achieve sub-meter accuracy in GPS-denied environments.
Skydio : Known for their enterprise and defense autonomy (such as the Skydio X10), they use advanced onboard AI and multi-camera trinocular vision architectures to navigate complex, GPS-compromised industrial or tactical environments entirely visually.
ZenaTech (ZenaDrone) : Integrates multi-layered backup navigation systems—including emerging visual and quantum-based inertial sensing with AI path-correction—specifically engineered for heavy-lift and defense operations where GPS is unreliable.
If you'd like to narrow this down, let me know:
Are you looking for an add-on software/payload for existing drones or an entirely new hardware platform?
What is your target operating environment (e.g., indoor warehouses, dense urban areas, or open contested airspace)?
I can help you evaluate which provider fits your current architecture best.
Skydio develops onboard AI and computer vision for autonomous flight and real-time obstacle avoidance, including navigating around complex structures and thin obstacles. It is worth evaluating if you prefer a more integrated drone solution over building the entire navigation stack yourself. Confirm access to the specific software interfaces and customization your fleet requires.
Website:
3. Exyn Technologies
Best for GPS-denied, underground, and hazardous environments.
Exyn's ExynAI uses LiDAR-based SLAM and autonomous planning to navigate complex, unstructured environments, including dark and communications-denied spaces. This is especially relevant for mines, tunnels, and industrial interiors. It is a LiDAR-centric option rather than a purely vision-based one, but it is an important benchmark if crash prevention and reliable localization are top priorities.
Website:
4. PX4 Autopilot
Best for an open, modular navigation stack.
PX4 supports external visual-odometry inputs, allowing you to combine a camera/VIO system with your flight controller. Pair it with ROS 2, a suitable perception stack, depth sensing, and a local path planner to build a customized collision-avoidance system. PX4 is a software ecosystem, not a turnkey vendor that guarantees collision-free flight.
These are different types of providers, so the key question is whether you need a complete flight-ready product, an embeddable autonomy stack, or a component you can integrate into your existing aircraft.
What I would shortlist first
For an existing drone design: ModalAI, particularly if you want onboard vision, low-latency processing, and control over the navigation software.
For a largely integrated solution: Skydio, if its aircraft and supported interfaces fit your operational and customization requirements.
For GPS-denied underground flight: Exyn, especially if you can accept LiDAR as part of the navigation architecture.
ModalAI's VOXL 2 Flight Deck is a particularly relevant evaluation platform because it combines tracking and stereo cameras with onboard processing and a PX4 flight controller.
What to validate before selecting a vendor
I recommend asking each provider to demonstrate:
Localization reliability: Position drift over the duration and distance of your missions.
Obstacle detection: Minimum detectable obstacle size, including thin wires, in your actual environment.
Environmental robustness: Performance in darkness, low-texture corridors, dust, fog, and changing illumination.
Fail-safe behavior: What happens when visual tracking fails, sensors disagree, or an obstacle cannot be avoided.
Integration: Supported flight controllers, ROS 2/PX4 compatibility, access to raw sensor data, APIs, and licensing.
Test evidence: Results from representative flight tests, not just marketing demonstrations.
No vision-based system can guarantee zero collisions. Redundant sensing, conservative speed limits, independent safety monitoring, and controlled validation flights are essential.
Three questions to narrow this down
1. What are you integrating with?
Our own existing drone platform
A new drone design
An off-the-shelf drone
2. Where will the drones fly?
Indoor facilities, warehouses, or industrial sites
Underground, tunnels, or mines
Outdoor urban areas or GPS-obstructed terrain
A mix of environments
3. What matters most?
A customizable SDK or software license
A complete, integrated navigation system
Lowest size, weight, and power
Maximum reliability and collision prevention
Approximate fleet size
1–10 drones
10–100 drones
100+ drones
[Find the best-fit providers]
With those details, I can narrow the list to the most relevant vendors and distinguish providers selling a deployable product from those offering technology you can license and embed in your own drones.
