Hesai Technology LiDAR Sensors: Deep Technical Research Report

Last updated: 2026-03-22

1. Company Overview

Background

Hesai Technology (NASDAQ: HSAI, HKEX: 02525.HK) is a Shanghai-headquartered LiDAR sensor company founded in 2014 by Yifan "David" Li, Kai Sun, and Shaoqing Xiang. The company designs and manufactures LiDAR products for ADAS, autonomous driving, robotics, and industrial applications. Hesai operates offices in Shanghai, Palo Alto, and Stuttgart, with customers spanning more than 40 countries and a workforce of 1,000-5,000 employees.

IPO and Capital Markets

NASDAQ IPO: February 2023, raising US$190 million
Hong Kong Secondary Listing: September 2025, raising US$614 million (HKEX: 02525.HK)
Market Capitalization: ~US$3.0 billion (as of early 2026)
Enterprise Value: ~US$2.1 billion

Unit Shipments and Scale

Metric	Value
Cumulative deliveries (all-time)	2,000,000+ units
2025 full-year deliveries	~1,600,000 units
2025 ADAS units	~1,400,000
2025 Robotics units	200,000+
Q3 2025 quarterly units	441,398
AT128 cumulative (as of Q2 2024)	380,000+ units
Monthly peak production	200,000+ units
Production cycle time	10 seconds per unit

Hesai became the world's first LiDAR pure-play to achieve profitability (2024) and the first to surpass 1 million cumulative units produced (2025). Full-year 2025 represented the fifth consecutive year of doubled annual shipments.

Financial Performance (TTM, early 2026)

Metric	Value
Revenue	US$380.9M
Gross Margin	41.4%
Operating Margin	6.3%
Net Margin	-3.7%
Total Cash	US$360.6M
Debt/Equity	0.11
Q3 2025 GAAP Net Income	RMB 256M (record)

Market Position

Global ADAS LiDAR revenue share: ~33% (2024)
Long-range ADAS segment: ~40% market share
Robotaxi LiDAR segment: 60-70% market share
Design wins: 24 automotive OEMs, 120+ vehicle models
OEM partners: Li Auto, Xiaomi, Changan, Geely, Zeekr, Great Wall Motors, BYD, and five major joint ventures (Volkswagen, GM, Audi, Toyota, Ford)
Robotics partners: Pony.ai, Hello Inc, JD Logistics, DREAME, Vbot, MOVIN
IP Leadership: Largest patent portfolio among LiDAR pure-play companies as of 2025

2. Full Product Line

Hesai's product portfolio spans ADAS, autonomous driving, robotics, and industrial applications. Below is the complete current product matrix.

2.1 ADAS / Automotive Long-Range

AT128 -- Automotive-Grade 120-deg Long-Range LiDAR

The AT128 is Hesai's mass-market ADAS LiDAR and the company's first integrated-architecture product. It was the volume workhorse through 2024-2025.

Parameter	Value
Channels	128
Pixel Resolution	1200 x 128
FOV	120 deg (H) x 25.4 deg (V)
Angular Resolution	0.1 deg (H) x 0.2 deg (V)
Range (@10% reflectivity)	210 m
Point Rate (single return)	1,536,000 pts/s
Power Consumption	13.5 W
Dimensions	136 x 114 x 49 mm (W x D x H)
Weight	940 g
Eye Safety	Class 1
Wavelength	905 nm
IP Rating	IP6K7 / IP6K9K
Functional Safety	ISO 26262 ASIL B product certification
Cybersecurity	ISO 21434 compliant process
Designed Lifetime	30,000+ hours

Reliability validation: UV aging, high-temperature, mechanical shock, salt spray, thermal shock with water splash, low-temperature wakeup, humid heat cycling, vibration with thermal cycling, IP6K7/IP6K9K dust resistance.

Key deployments: Li Auto L-series, Pony.ai 7th-gen robotaxis (4x AT128 per vehicle), JiDU, HiPhi, Lotus.

ATX -- Next-Generation Compact 120-deg Long-Range LiDAR

The ATX is the successor to the AT128, featuring the FMC500 SoC and Photon Isolation technology. Mass production deliveries begin April 2026.

Parameter	ATX	ATXP (Flagship)
Channels	192	256
FOV	120 deg (H) x 18.3 deg (V)	120 deg (H) x 20 deg (V)
Angular Resolution	0.08 deg (H) x 0.1 deg (V)	0.08 deg (H) x 0.05 deg (V)
Range (@10% reflectivity)	200 m	230 m
Point Rate (single return)	1,740,000 pts/s	3,840,000 pts/s
Power Consumption	10 W	8 W
Dimensions	102 x 102 x 30 mm (W x D x H)	102 x 102 x 30 mm
Weight	360 g	360 g
IP Rating	IP6K7 / IP6K9K	IP6K7 / IP6K9K
Functional Safety	ISO 26262 ASIL B	ISO 26262 ASIL B
Noise Filtering	99.9% (IPE)	99.9% (IPE)

Key improvements over AT128: 60% smaller by volume, 25mm minimum window height, equipped with Hesai's Fermi C500 SoC, built-in rain/fog identification with pixel-level marking. Order backlog exceeds 4 million units.

ET25 -- Ultra-Thin Behind-Windshield Long-Range LiDAR

Parameter	Value
FOV	120 deg (H) x 25 deg (V)
Angular Resolution	0.05 deg (H) x 0.05 deg (V)
Range (without windshield)	250 m (@10% reflectivity)
Range (behind windshield)	225 m
Point Rate	3,000,000+ pts/s
Power Consumption	12 W
Height	25 mm
Wavelength	905 nm
Eye Safety	Class 1

Designed for in-cabin placement behind the windshield. FAW Hongqi adopted the ET25 for production vehicles.

ETX -- Ultra-Long-Range Behind-Windshield LiDAR (Next-Gen L3)

The ETX is Hesai's next-generation flagship long-range sensor, featuring 800 channels and the world's first fully in-house digital single-photon platform.

Parameter	Value
Channels	800
FOV	120 deg (H) x 20 deg (V)
Angular Resolution	0.05 deg (H) x 0.025 deg (V)
Range (@10% reflectivity)	400 m
Point Rate (single return)	5,600,000 pts/s
Resolution vs AT128	8x higher
Power Consumption	11 W
Dimensions	137 x 120 x 36 mm (window height 32 mm)
Weight	600 g
Operating Temperature	-40 deg C to +85 deg C
Noise Level	<= 25 dB(A)
IP Rating	IP6K7 / IP6K9K
Eye Safety	Class 1
Standards	ISO 26262, ISO 21434, ISO 21448 compliant

Mass production planned for late 2026 or early 2027. First passenger-vehicle series-production program secured.

Parameter	Value
Architecture	Fully solid-state (no moving parts)
FOV	100 deg (H) x 75 deg (V)
Pixel Resolution	160 (H) x 120 (V)
Angular Resolution	0.625 deg (H) x 0.625 deg (V)
Range	100 m max
Point Rate (single return)	192,000 pts/s
Point Rate (dual return)	384,000 pts/s
Frame Rate	10 Hz
Minimum Window Size	70 x 50 mm
Wavelength	905 nm
Eye Safety	Class 1

Designed for fender, grille, or bumper integration. Targets ADAS blind-spot coverage.

FTX -- Next-Generation Fully Solid-State Short-Range LiDAR

Parameter	FTX-180	FTX-140
FOV	180 deg (H) x 140 deg (V)	140 deg (H) x 105 deg (V)
Angular Resolution	0.62 deg (finest)	0.56 deg (H) x 0.56 deg (V)
Pixel Resolution	216 x 192	256 x 192
Range (@10% reflectivity)	20 m	30 m
Point Rate (single return)	430,000 pts/s	492,000 pts/s
Scanning Frequency	10 Hz	10 Hz
Dimensions	55 x 60 x 38.7 mm	55 x 60 x 32 mm
Weight	~160 g	~160 g
Power Consumption	<6 W	<6 W
IP Rating	IP6K7 / IP6K9K	IP6K7 / IP6K9K
Eye Safety	Class 1	Class 1

The FTX holds the world's largest FOV for an automotive-grade solid-state LiDAR (180 x 140 deg). 66% lighter than the FT120. Pairs with ETX as Hesai's "Infinity Eye B" L3 perception solution. Mass production planned for 2026.

QT128 -- 360-deg Ultra-Wide-View Short-Range LiDAR

Parameter	Value
Channels	128
FOV	360 deg (H) x 105 deg (V)
Angular Resolution	0.4 deg (H) x 0.4 deg (V)
Range (@10% reflectivity)	20 m
Range Accuracy	+/-3 cm
Point Rate (single return)	1,152,000 pts/s
Dimensions	87 mm diameter x 83.9 mm height
Weight	~700 g
Power Consumption	12 W
Operating Temperature	-40 deg C to +85 deg C
IP Rating	IP6K7 / IP6K9K
Functional Safety	ISO 26262 ASIL B
Eye Safety	Class 1

Passed 50+ DV tests per international OEM standards. Provides high-quality reflectivity data for enhanced object recognition.

2.3 Mid-Range Mechanical (360-deg Spinning)

XT32 -- 32-Channel High-Precision Mid-Range LiDAR

Parameter	Value
Channels	32
FOV	360 deg (H) x 31 deg (V)
Angular Resolution	0.18 deg (H) x 1.0 deg (V)
Range	0.05 - 120 m
Range Precision	0.5 cm (1-sigma)
Range Accuracy	+/-1 cm
Point Rate (single return)	640,000 pts/s
Point Rate (dual return)	1,280,000 pts/s
Frame Rate	5 / 10 / 20 Hz
Return Modes	Last, Strongest, Dual
Wavelength	905 nm
Dimensions	76 mm diameter x 103.2 mm height
Weight	800 g
Power Consumption	<30 W peak (typical ~15 W)
Operating Temperature	-20 deg C to +60 deg C
Interface	100 Mbps Ethernet (UDP/IP)
IP Rating	IP6K7
Eye Safety	Class 1
Designed Lifetime	30,000+ hours
Minimum Range	0.05 m (effectively zero)

The XT32 is widely used for drone LiDAR, mobile mapping, surveying, precision agriculture, mining, and AV development. Continuously outputs valid point cloud data even when objects directly touch the sensor.

XT32M (XT32M2X) -- Extended Mid-Range Variant

Parameter	Value
Channels	32
FOV	360 deg (H) x 40.3 deg (V)
Angular Resolution	0.09 deg (H) x 1.3 deg (V)
Range (@10% reflectivity)	80 m (300 m max)
Range Precision	0.5 cm (1-sigma)
Range Accuracy	+/-1 cm
Point Rate (single return)	640,000 pts/s
Point Rate (dual return)	1,280,000 pts/s
Point Rate (triple return)	1,920,000 pts/s
Frame Rate	5 / 10 / 20 Hz
Wavelength	905 nm
Power Consumption	10 W
Operating Temperature	-20 deg C to +60 deg C
Dimensions	75 mm diameter x 93 mm height
Weight	490 g
Eye Safety	Class 1

XT16 -- 16-Channel Economy Mid-Range

Parameter	Value
Channels	16
FOV	360 deg (H) x 30 deg (V)
Angular Resolution	0.18 deg (H) x 2.0 deg (V)
Range	0.05 - 120 m
Point Rate (single return)	320,000 pts/s
Dimensions	76 mm diameter x 103.2 mm height
Weight	800 g

2.4 360-deg Long-Range (Robotaxi / L4)

OT128 -- Automotive-Grade 360-deg Long-Range LiDAR

Parameter	Value
Channels	128
FOV	360 deg (H) x 40 deg (V)
Angular Resolution	0.1 deg (H) x 0.125 deg (V)
Range (@10% reflectivity)	200 m (230 m max)
Range Accuracy	+/-3 cm
Point Rate (single return)	3,456,000 pts/s
Point Rate (dual return)	6,912,000 pts/s
Power Consumption	29 W
Dimensions	118 mm diameter x 132.3 mm height
IP Rating	IP6K7 / IP6K9K
Designed Lifetime	30,000+ hours (3x industry average)
Eye Safety	Class 1
Standards	ISO 26262, ISO 21434 compliant

66% fewer parts than predecessor. Production time reduced by 95% via automated manufacturing. Built-in Intelligent Point Cloud Engine (IPE) for rain/fog/exhaust/water splash detection with pixel-level marking.

Pandar128 (Legacy)

Parameter	Value
Channels	128
Designed Lifetime	30,000+ hours
Certification	ASIL B

The Pandar128 was Hesai's flagship 360-deg spinning LiDAR prior to the OT128. Still supported but effectively superseded.

2.5 Robotics

JT16 -- Mini 360-deg 3D LiDAR for Robotics

Parameter	Value
FOV	360 deg (H) x 40 deg (V)
Angular Resolution	0.6 deg (H) x 2.67 deg (V)
Range (@10% reflectivity)	30 m (100 m max)
Scanning Frequency	5 / 10 Hz
Power Consumption	4.3 W
Dimensions	55 mm diameter x 64 mm height
Weight	<200 g
Built-in IMU	Yes
IP Rating	IP6K7 / IP6K9K
Eye Safety	Class 1

70% smaller in volume than similar products. Built-in waveform processing for cover lens contamination detection. Cumulative deliveries exceeded 200,000 units. Used in lawn-mowing robots, humanoid robots, AMRs, and 3D digitalization devices.

JT128 -- Higher-Resolution Robotics LiDAR

128-channel variant of the JT series for robotics applications requiring denser point clouds. Range up to 60 m.

2.6 Legacy Products

Product	Channels	Range	Status
Pandar40P	40	200 m (@10%)	Supported, effectively superseded
Pandar40M	40	—	Supported, effectively superseded
Pandar64	64	200 m (@10%)	Supported, effectively superseded
Pandar20A/20B	20	—	Supported, effectively superseded
PandarQT	64	—	Superseded by QT128

3. XT32 Deep Specifications

The XT32 (originally launched as PandarXT-32) is particularly relevant for AV development, drone mapping, and mobile sensing platforms. This section consolidates all known specifications.

3.1 Optical and Sensing

Parameter	Value
Laser Wavelength	905 nm
Channels	32
Eye Safety	Class 1 (IEC 60825-1)
Ranging Principle	Time-of-Flight (ToF)
Measurement Range	0.05 m to 120 m
Minimum Detection Range	0.05 m (near-zero; valid data output on contact)
Range Precision	0.5 cm (1-sigma, typical)
Range Accuracy	+/-1 cm

3.2 Field of View and Resolution

Parameter	Value
Horizontal FOV	360 deg
Vertical FOV	31 deg
Horizontal Angular Resolution	0.18 deg
Vertical Angular Resolution	1.0 deg
Vertical Channel Distribution	Non-uniform (denser near horizon)

3.3 Data Output

Parameter	Value
Point Rate (single return)	640,000 pts/s
Point Rate (dual return)	1,280,000 pts/s
Frame Rate Options	5 Hz, 10 Hz, 20 Hz
Default Spin Rate	600 RPM (= 10 Hz)
Return Modes	Last Return, Strongest Return, Dual Return
Data Interface	100 Mbps Ethernet (UDP/IP)
Output Packets	Point Cloud Data Packets + GPS Data Packets
GPS Synchronization	Supported (PPS + NMEA)

3.4 Mechanical and Electrical

Parameter	Value
Dimensions	76 mm diameter x 103.2 mm height
Weight	800 g
Power Consumption (peak)	<30 W (all operating conditions)
Power Consumption (typical)	~15 W (may sustain 15 W in cold-start below 0 deg C)
Operating Temperature	-20 deg C to +60 deg C
Storage Temperature	-40 deg C to +85 deg C (typical for Hesai products)
IP Rating	IP6K7
Connector	Ethernet + power multi-pin

3.5 ASIC Integration

The XT32 was the first Hesai product to feature self-developed proprietary LiDAR ASICs, integrating laser driver, analog front-end, and waveform processing into custom silicon. This was a key differentiator at launch that reduced BOM cost and improved signal-to-noise ratio.

3.6 Use Cases

Airport airside autonomous vehicle development and testing
Drone-based aerial surveying and mapping
Mobile laser scanning (MLS)
Construction volumetrics and site surveying
Precision agriculture and forestry
Mining operations
Robotics development platforms
Ground-truth data collection

4. AT128 ASIL-B Certification

ISO 26262 Functional Safety

The AT128 has achieved ISO 26262 ASIL B (Automotive Safety Integrity Level B) product certification, verified by SGS-TUV. This covers:

Product-level certification: The AT128 hardware and firmware meet ASIL B requirements for systematic and random hardware faults
Development process compliance: The development process follows ISO 26262 requirements from concept through production
ISO 21434 cybersecurity: Compliant development process for automotive cybersecurity

What ASIL B Means for Integration

ASIL B is the second level of the four-tier automotive safety classification (A through D). For LiDAR sensors in ADAS applications:

Suitable for L2/L2+ ADAS functions where the LiDAR provides redundant perception alongside cameras and radar
Supports Highway Pilot, Traffic Jam Pilot, and similar highway-centric functions
Appropriate for sensor fusion architectures where no single sensor is the sole safety-critical path
For L3+ applications requiring ASIL D system-level compliance, the LiDAR at ASIL B can be paired with ASIL B camera/radar in a complementary decomposition

Other Hesai Products with ASIL B

ATX: ISO 26262 ASIL B certified (via SGS-TUV)
QT128: ISO 26262 ASIL B certified
FMC500 chip: World's first LiDAR SoC with on-chip functional safety and cybersecurity certification

5. Core Technology

5.1 Wavelength: 905 nm

All Hesai production LiDAR sensors use 905 nm laser wavelength. This is a deliberate architectural choice with the following tradeoffs:

Advantages of 905 nm:

Mature semiconductor laser technology with high electro-optical conversion efficiency
Lower component cost (silicon-based detectors, standard semiconductor processes)
Proven supply chain for high-volume automotive manufacturing
Less fiber coupling and amplification required versus 1550 nm fiber lasers
Compatible with both APD and SiPM/SPAD detector technologies

Tradeoffs vs 1550 nm:

1550 nm allows ~40x higher laser power within eye safety limits (IEC 60825-1)
1550 nm exhibits less atmospheric scattering and absorption
1550 nm is better suited for FMCW architectures due to greater coherence
However, 1550 nm requires expensive InGaAs detectors and fiber laser sources
1550 nm manufacturing is significantly more complex for volume production

Hesai's position is that 905 nm, combined with advanced SiPM detectors and proprietary signal processing (FMC500 chip), achieves competitive range (200-400 m @10%) while maintaining cost and manufacturing advantages critical for mass-market ADAS deployment.

5.2 Detection Technology: Time-of-Flight (ToF)

All current Hesai products use direct Time-of-Flight measurement:

Laser pulse emitted, reflected off target, return time measured
Distance = (speed of light x round-trip time) / 2
Hesai uses single-photon detectors (SiPM/SPAD arrays) that generate effective data signals with just a few photons
Custom ASICs (4 generations) handle waveform processing and timing

ToF vs FMCW:

Characteristic	ToF (Hesai)	FMCW
Measurement	Distance only	Distance + instantaneous velocity
Ambient light immunity	Moderate (improved by IPE)	High (coherent detection)
Signal processing	Simpler	More complex
Manufacturing maturity	Production-ready, millions shipped	Pre-production / early production
Cost	Lower	Higher
Anti-interference	Good (with Photon Isolation)	Excellent (coherent detection)

5.3 Proprietary Technologies

Intelligent Point Cloud Engine (IPE)

The IPE is Hesai's on-chip signal processing pipeline that:

Filters out 99.9% of environmental noise (rain, fog, dust, exhaust)
Detects and marks rain, fog, exhaust fumes, and water splashes at pixel level
Enables real-time environmental classification within the sensor itself
Runs on the FMC500 SoC

Photon Isolation Technology

Ensures that photons received by each laser channel do not interfere with adjacent channels. Key properties:

Channel count is 10x that of traditional SPAD solutions
Suppresses "ghosting" artifacts
Enhances safety margins for autonomous driving applications
Integrated across ATX and ETX product lines

FMC500 SoC

Hesai's 4th-generation custom chip, built on RISC-V architecture:

Integrates MCU, FPGA, and ADC on a single die
256-core waveform processing engine
World's first LiDAR master control chip with dual functional safety + cybersecurity certification
Intelligent noise filtering adaptive to complex weather conditions
Powers the ATX and ETX product lines

5.4 Scanning Architectures

Architecture	Products	Mechanism
Hybrid solid-state (rotating prism)	AT128, ATX, ET25, ETX	Internal rotating mirror/prism, no external rotation
Mechanical spinning (360 deg)	XT32, XT16, OT128, QT128, JT16	Full rotary assembly
Fully solid-state (flash/OPA)	FT120, FTX	No moving parts, electronic beam steering

6. Point Cloud Format and ROS Driver

6.1 Point Cloud Data Structure

Hesai LiDAR sensors output point cloud data via UDP packets over Ethernet. The core point data structure is LidarPointXYZICRTT:

Field	Type	Description
x	float32	X coordinate (meters)
y	float32	Y coordinate (meters)
z	float32	Z coordinate (meters)
intensity	uint8/float32	Reflectivity / return intensity
channel	uint16	Laser channel ID (also called ring, laser_id)
ring	uint16	Vertical channel index
timestamp	float64	Per-point timestamp (seconds + nanoseconds)
return_type	uint8	Return type indicator (last, strongest, etc.)

Timestamp composition:

Absolute time = Date + Time (to the second) + Microseconds
Date and Time can come from the current Point Cloud Data Packet (6 bytes) or previous GPS Data Packet
Microseconds from current packet (4 bytes)
Per-point timestamps include packet timestamp + firing channel timing correction

6.2 ROS Driver: HesaiLidar_ROS_2.0

Repository: https://github.com/HesaiTechnology/HesaiLidar_ROS_2.0

Supported Platforms:

Ubuntu	ROS 1	ROS 2
16.04	Kinetic	—
18.04	Melodic	Dashing
20.04	Noetic	Foxy
22.04	—	Humble
24.04	—	Jazzy

Supported LiDAR Models:

Pandar series: Pandar40P, Pandar64, Pandar128E3X
OT series: OT128
QT series: PandarQT, QT128C2X
XT series: PandarXT, PandarXT-16, XT32M2X
AT series: AT128E2X, AT128P, ATX
FT series: FT120
JT series: JT16, JT128

Default ROS Topics:

Topic	Message Type	Description
`/lidar_points`	sensor_msgs/PointCloud2	3D point cloud
`/lidar_imu`	sensor_msgs/Imu	IMU data (if available)
`/lidar_packets`	—	Raw UDP packets
`/lidar_packets_loss`	—	Packet loss monitoring

Data Source Modes:

source_type	Description
1	Real-time UDP connection
2	PCAP file playback
3	Rosbag packet playback
4	Serial connection

Key Configuration Parameters:

device_ip_address: LiDAR device IP
udp_port: UDP destination port (default 2368)
ptc_port: PTC control port (default 9347)
use_gpu: GPU acceleration toggle
thread_num: Parser thread count
frame_frequency: Output frame rate
echo_mode_filter: Return mode filtering
Transform: x, y, z offsets + roll, pitch, yaw rotation

Installation:

bash

# Clone with submodules
git clone --recurse-submodules https://github.com/HesaiTechnology/HesaiLidar_ROS_2.0.git

# Dependencies
sudo apt install libboost-all-dev libyaml-cpp-dev

# ROS 2 build
colcon build --symlink-install
. install/local_setup.bash
ros2 launch hesai_ros_driver start.py

Multi-LiDAR fusion is supported through multiple driver instances with configurable topic names per sensor.

6.3 SDK: HesaiLidar_SDK_2.0

Repository: https://github.com/HesaiTechnology/HesaiLidar_SDK_2.0

Language: C++ (89.1%), with CUDA support (9.1%)
Platforms: Ubuntu 16/18/20/22.04, Windows 10
Build: CMake 3.8.0+, G++ 7.5+ or MSVC 2019+
Dependencies: PCL, libpcap, OpenSSL, libyaml-cpp

Capabilities:

Real-time online data parsing and PCAP offline parsing
Multi-LiDAR simultaneous processing with IP/port filtering
GPU-accelerated point cloud processing
PCD file export
Coordinate transformation utilities
Packet loss statistics and monitoring
TLS/mTLS encrypted PTC communication
Per-point precise timestamps with firing-channel correction

6.4 PandarView 2 Visualization Tool

Qt-based visualization software (fully self-developed, not Paraview-based)
Displays azimuth, distance, intensity, timestamp per selected point
Channel-level show/hide control
Colormap by azimuth, distance, elevation, channel, or timestamp
Available for Windows 10 and Ubuntu 16.04/18.04/20.04
Download from: https://www.hesaitech.com/downloads/

7. Adverse Weather Performance

7.1 Technical Foundations

Hesai's approach to adverse weather relies on three pillars:

Active Light Emission: LiDAR's optical power density reaching objects at 100 m is approximately 7x that of ambient visible light, enabling detection in conditions where cameras fail due to insufficient ambient illumination.

Large Aperture Optics: The AT-series rotating mirror design uses significantly larger optical apertures than MEMS-based systems. This reduces the proportional impact of water droplets on sensor windows. For cameras, a single water droplet can block a large area of the image; the larger LiDAR aperture is more resilient to partial occlusion.

Single-Photon Detection: Production LiDAR uses SiPM detectors that generate valid data signals from just a few returned photons, maintaining sensing consistency across varying weather conditions.

7.2 Intelligent Point Cloud Engine (IPE) Weather Processing

The IPE, integrated into the FMC500 SoC and deployed across ATX, ETX, and OT128 products, provides:

Filtering of 99.9% of environmental noise
Real-time pixel-level classification of rain, fog, exhaust fumes, and road water splashes
Adaptive noise filtering using the 256-core waveform processing engine

7.3 Performance by Condition

Fog:

Point clouds clearly show vehicles at 50 m when cameras cannot identify them
Valid data points far exceed algorithmic detection requirements
Mid-to-close range performance in mild/moderate fog is nearly equivalent to clear-day performance

Rain:

Reliable detection of vehicles and pedestrians maintained
Ranging capability somewhat reduced in moderate rain
Performance at mid-to-close range remains strong

General Degradation Characteristics:

A LiDAR with 200 m range (@10% reflectivity) in clear conditions shows approximately 33% better weather penetration performance than a 150 m range system
Even with reduced ranging capability, data loss does not occur -- the system continues to detect vehicles and pedestrians

7.4 Environmental Protection

All automotive-grade Hesai products are rated IP6K7 and IP6K9K, providing:

Complete dust ingress protection (6K)
Protection against temporary immersion (7) and high-pressure/high-temperature water jets (9K)

8. Competitive Comparison

8.1 Hesai vs RoboSense

Attribute	Hesai	RoboSense
Headquarters	Shanghai, China	Shenzhen, China
Public Listing	NASDAQ + HKEX	HKEX (02522.HK)
Wavelength	905 nm	905 nm
Primary ADAS Product	ATX (successor to AT128)	M2 (successor to M1 Plus)
ADAS Range (@10%)	200-230 m (ATX)	200-250 m (M2)
ADAS Resolution	0.08 x 0.05 deg (ATXP)	0.1 x 0.1 deg (M2 ROI)
ADAS Point Rate	3,840,000 pts/s (ATXP)	—
ADAS Price	~$200 (ATX)	~$200 (estimated)
Functional Safety	ASIL B (AT128, ATX, QT128)	ASIL B (M platform)
Market Share (ADAS)	~33-40%	#2 globally
Cumulative Shipments	2,000,000+	—
Scanning Architecture	Rotating prism (ADAS)	MEMS-based (M platform)
Primary Focus	Robotaxi + passenger ADAS	Passenger ADAS + low-speed AV

8.2 Hesai vs Ouster (Velodyne merged into Ouster, 2023)

Attribute	Hesai	Ouster
Headquarters	Shanghai	San Francisco
Wavelength	905 nm	865 nm (VCSEL array)
Detection Technology	SiPM ToF	SPAD array digital ToF
Key Products	XT32, OT128, ATX	OS1 (mid), OS2 (long)
Mid-Range Channels	32 (XT32)	32/64/128 (OS1)
Mid-Range (@10%)	120 m (XT32)	90 m (OS1-128)
Long-Range (@10%)	200 m (OT128)	200 m (OS2)
Point Rate	6,912,000 (OT128 dual)	5,200,000 (OS1-128)
IP Rating	IP6K7 / IP6K9K	IP68 / IP69K
Price Point	Lower (volume advantage)	Higher
Automotive ASIL	ASIL B certified	Limited automotive certification
Volume Scale	Millions per year	Thousands per year
Market Focus	Auto ADAS + robotaxi	Multi-vertical (security, mapping, AV)

Ouster's digital LiDAR architecture produces camera-like near-IR imagery alongside 3D point clouds, a differentiator for certain applications. However, Ouster's IP position has been overtaken by faster-growing Chinese companies in the patent landscape.

8.3 Hesai vs Luminar

Attribute	Hesai	Luminar
Headquarters	Shanghai	Orlando, FL
Wavelength	905 nm	1550 nm
Detection Technology	SiPM/SPAD ToF	InGaAs APD ToF
Key ADAS Product	ATX	Iris / Iris+
Max Range	400 m (ETX @10%)	600 m (max) / 250 m (dark objects)
Eye Safety Power Budget	Standard (905 nm Class 1)	17-40x higher power per eye safety
Resolution	0.05 x 0.025 deg (ETX)	Up to 300 pts/sq deg
Price	~$200 (ATX mass market)	Sub-$1,000 target
Volume Production	Millions per year	Ramp-up phase
OEM Partners	24 OEMs, 120+ models	Volvo, Mercedes-Benz, SAIC
Profitability	Profitable (2024)	Pre-profit
Weather Penetration	IPE noise filtering	1550 nm inherent advantage
Manufacturing	Highly automated, 10s cycle	Lower volume

Luminar's 1550 nm wavelength allows significantly higher laser power within eye safety limits, providing theoretical advantages in range and adverse weather. However, Hesai's 905 nm platform achieves competitive real-world performance through advanced signal processing while maintaining dramatic cost and volume advantages.

8.4 Summary Comparison Matrix

Metric	Hesai ATX	RoboSense M2	Ouster OS2	Luminar Iris
Wavelength	905 nm	905 nm	865 nm	1550 nm
Range @10%	200-230 m	200-250 m	200 m	250 m
ADAS ASIL	B	B	—	B (target)
Price (est.)	~$200	~$200	$600-1,200	<$1,000
Volume Scale	Millions	Hundreds of thousands	Thousands	Thousands
Profitability	Yes	Improving	Improving	No

9. SDK and Software Tools

9.1 Complete Tool Ecosystem

Tool	Purpose	Platform
HesaiLidar_SDK_2.0	C++ SDK for point cloud parsing, GPU acceleration	Ubuntu, Windows
HesaiLidar_ROS_2.0	ROS/ROS2 driver for all current models	Ubuntu (ROS Kinetic through Jazzy)
HesaiLidar_General_ROS	Legacy ROS driver for Pandar-era products	Ubuntu (ROS Kinetic through Noetic)
PandarView 2	Qt-based point cloud visualization	Windows 10, Ubuntu 16/18/20
PandarView	ParaView-based visualization (legacy)	Windows, Ubuntu

9.2 SDK Architecture

The SDK 2.0 provides a modular C++ framework:

Driver module: UDP packet reception and parsing
Library module: Point cloud data structures and transformations
Tool module: PCD export, visualization utilities
Test module: Unit testing framework

Communication protocols:

UDP for point cloud data packets
PTC (Pandar TCP Control) for device configuration
Optional TLS/mTLS encryption for PTC commands
Configurable timeouts for initialization and connection

9.3 Integration with Third-Party Frameworks

Autoware: Native support via PointCloud2 message format
Apollo: Compatible through UDP packet interface
NVIDIA DriveWorks: Integration path via SDK
PCL (Point Cloud Library): Direct PCD export support
PCAP: Full offline playback support for development and testing

10. Reliability and MTBF Data

10.1 Design Lifetime

Product	Designed Lifetime
OT128	30,000+ hours (3x industry average)
Pandar128	30,000+ hours
XT32 series	30,000+ hours
AT128	Automotive-grade (30,000+ hours implied)

30,000 hours at continuous operation equals approximately 3.4 years. For typical automotive duty cycles (8-12 hours/day), this translates to 7-10 years of operational life.

10.2 Validation Testing

All automotive-grade Hesai products undergo 50+ design validation (DV) tests per internationally recognized OEM standards:

Test Category	Specific Tests
Thermal	High-temperature exposure, low-temperature wakeup, thermal shock with water splash, humid heat cycling
Mechanical	Vibration with thermal cycling, mechanical shock
Environmental	UV aging, salt spray corrosion, IP6K7 immersion, IP6K9K high-pressure spray
Electrical	Power cycling, ESD, EMC

10.3 Component-Level Qualification

All key components meet AEC-Q qualification standards (AEC-Q100 for ICs, AEC-Q101 for discretes)
FMC500 SoC carries on-chip functional safety certification
Operating temperature ranges extend to -40 deg C to +85 deg C for automotive-grade products

10.4 MTBF

Specific MTBF (Mean Time Between Failure) figures are not publicly disclosed by Hesai. The 30,000+ hour designed lifetime provides a lower-bound estimate. For automotive qualification, OEMs typically require MTBF figures in the hundreds of thousands of hours, which would be shared under NDA during the design-win qualification process.

11. Cost and Pricing

11.1 Current Pricing (Estimated, 2026)

Product	Estimated Unit Price	Segment
ATX (ADAS)	~$200	Mass-market passenger vehicles
AT128 (ADAS)	~$400 (pre-ATX transition)	Mid-to-high-end ADAS
OT128 (360-deg)	$1,000-3,000 (est.)	Robotaxi / L4
XT32 (mapping)	$2,000-4,000 (est.)	Development / mapping
JT16 (robotics)	Low hundreds (est.)	Consumer robotics
FT120 (blind spot)	Sub-$200 (est.)	ADAS supplement

11.2 Pricing Trends

Hesai announced plans to halve LiDAR prices in 2025, with the ATX at ~$200 representing roughly half the AT128 price
The $200 price point is identified as the threshold where LiDAR becomes economically viable for vehicles in the RMB 100,000 (~$14,000) price range
Expected 2026 content per vehicle: $500-1,000 for multi-LiDAR configurations (1x long-range + 2-3x blind-spot)
Highly automated production (10-second cycle time) is the primary cost reduction lever

11.3 Cost Structure

Gross margin: 41.4% (TTM early 2026), indicating significant room above BOM cost
Scale economics: Production capacity doubling from 2 million to 4 million units in 2026
Vertical integration: 4 generations of custom ASICs reduce dependence on expensive off-the-shelf components
FMC500 SoC: Integrates MCU + FPGA + ADC onto a single die, reducing component count and assembly cost

12. Future Roadmap

12.1 Near-Term (2026)

Initiative	Timeline	Details
ATX mass production	April 2026	4M+ unit order backlog, 256-line flagship with FMC500
Production capacity doubling	2026	2M to 4M+ annual capacity
ETX + FTX L3 suite	Late 2026 / early 2027	800-channel long-range + 180-deg solid-state blind-spot
Bangkok factory	Early 2027 operations	New manufacturing facility in Thailand

12.2 Product Evolution

L3 Autonomous Driving: The "Infinity Eye" solution pairs ETX (long-range, behind-windshield) with FTX (short-range, blind-spot) for panoramic L3/L4 perception. First series-production program secured.

Physical AI / Robotics: Expansion into humanoid robots, lawn-mowing robots, delivery drones (Meituan's Keeta), autonomous mobile robots. JT-series deliveries exceeding 200,000 units with growing momentum.

NVIDIA Partnership: Collaboration on L4 autonomous driving rollout, integrating Hesai LiDAR with NVIDIA compute platforms.

12.3 Technology Trajectory

SiPM / Single-Photon Platform: The ETX's fully in-house digital single-photon platform (laser emission + single-photon detection + signal processing) represents Hesai's technology direction -- pushing 905 nm ToF performance to ranges previously associated with 1550 nm systems (400 m @10%).

FMCW: Hesai has not publicly announced FMCW product plans. The company's current strategy doubles down on advancing 905 nm ToF performance through silicon photonics and signal processing innovations. FMCW remains a technology to watch industry-wide, with companies like Aurora (FirstLight) and Aeva pursuing it, but commercialization timelines remain uncertain.

RISC-V SoC Roadmap: The FMC500 establishes Hesai's custom silicon platform on RISC-V architecture, providing a foundation for future generations with increasing integration, lower power, and enhanced processing capabilities.

12.4 Market Outlook

ADAS LiDAR shipments projected at 2-3 million units in 2026
China EV market LiDAR adoption rate at 28% and growing
LiDAR-equipped vehicles claimed to reduce fatal highway accidents by 90% vs camera-only
Automotive LiDAR market projected to reach US$25.75 billion by 2035
Hesai positioned as the clear volume leader with both cost and technology advantages

SLAM Methods

Methods

Hesai Technology LiDAR Sensors: Deep Technical Research Report ​

1. Company Overview ​

Background ​

IPO and Capital Markets ​

Unit Shipments and Scale ​

Financial Performance (TTM, early 2026) ​

Market Position ​

2. Full Product Line ​

2.1 ADAS / Automotive Long-Range ​

AT128 -- Automotive-Grade 120-deg Long-Range LiDAR ​

ATX -- Next-Generation Compact 120-deg Long-Range LiDAR ​

ET25 -- Ultra-Thin Behind-Windshield Long-Range LiDAR ​

ETX -- Ultra-Long-Range Behind-Windshield LiDAR (Next-Gen L3) ​

2.2 Short-Range / Blind-Spot Detection ​

FT120 -- Fully Solid-State Blind-Spot LiDAR ​

FTX -- Next-Generation Fully Solid-State Short-Range LiDAR ​

QT128 -- 360-deg Ultra-Wide-View Short-Range LiDAR ​

2.3 Mid-Range Mechanical (360-deg Spinning) ​

XT32 -- 32-Channel High-Precision Mid-Range LiDAR ​

XT32M (XT32M2X) -- Extended Mid-Range Variant ​

XT16 -- 16-Channel Economy Mid-Range ​

2.4 360-deg Long-Range (Robotaxi / L4) ​

OT128 -- Automotive-Grade 360-deg Long-Range LiDAR ​

Pandar128 (Legacy) ​

2.5 Robotics ​

JT16 -- Mini 360-deg 3D LiDAR for Robotics ​

JT128 -- Higher-Resolution Robotics LiDAR ​

2.6 Legacy Products ​

3. XT32 Deep Specifications ​