Industry Research

Appearance

Cruise Autonomous Vehicle Division: Exhaustive Technical Writeup ​

Last Updated: March 2026

Table of Contents ​

  1. Company Overview
  2. Vehicle Platform
  3. Sensor Suite
  4. Onboard Compute
  5. Autonomy Software Stack
  6. Machine Learning & AI
  7. Mapping & Localization
  8. Simulation Platform
  9. Cloud & Data Infrastructure
  10. Programming Languages & Tools
  11. Safety & Redundancy
  12. Fleet Operations
  13. Regulatory & Safety Record
  14. Key Partnerships
  15. Research & Publications
  16. Current Status (2025-2026)

1. Company Overview ​

Founding & History ​

Cruise was founded in 2013 by Kyle Vogt and Dan Kan in San Francisco, California. The company participated in Y Combinator as part of its startup accelerator program. Cruise's first product was the RP-1, announced in June 2014 -- a $10,000 aftermarket retrofit kit designed for 2012+ Audi A4 and S4 vehicles that would enable limited highway autonomy. The RP-1 was abandoned in January 2014 in favor of pursuing full autonomy, initially using the Nissan Leaf as a development platform.

In March 2016, General Motors acquired Cruise Automation for approximately $1 billion. At the time of acquisition, Cruise had roughly 40 employees and had raised $20 million in venture funding.

Key Leaders ​

LeaderRoleTenure
Kyle VogtCo-founder; CEO (interim Dec 2021, permanent Feb 2022)2013 -- Nov 2023 (resigned)
Dan KanCo-founder2013 -- departed post-acquisition
Dan AmmannCEO (former GM President)Nov 2018 -- Dec 2021 (departed)
Mo ElshenawyEVP Engineering / President & CTO; co-president post-VogtPre-2023 -- Apr 2025 (transition)
Marc WhittenCEO (former Amazon/Microsoft exec)Jul 2024 -- Feb 2025 (departed in GM absorption)
Craig GliddenCo-president (post-Vogt, GM General Counsel)Nov 2023 -- 2024
Carl JenkinsVP Hardware2018 -- onward
Sterling AndersonHired post-Cruise (former Tesla Autopilot chief)2025 -- present (GM autonomous driving)

Funding & Investment ​

GM's cumulative spending on Cruise reached approximately $12.1 billion:

YearInvestment
2016$1.0B (acquisition)
2017$600M
2018$700M
2019$1.0B
2020$900M
2021$1.2B
2022$1.9B
2023$2.7B (peak)
2024$1.7B
2025$400M
Total~$12.1B

Major external funding rounds:

  • May 2018: Honda invested $750M equity + committed ~$2B over 12 years ($2.75B total)
  • May 2019: SoftBank Vision Fund invested $2.25B; total valuation reached $19B
  • January 2021: Microsoft, Honda, and institutional investors contributed $2B; valuation reached $30B
  • April 2021: Walmart joined a $2.75B round alongside Microsoft and Honda
  • March 2022: GM acquired SoftBank Vision Fund 1's equity for $2.1B and invested an additional $1.35B

Peak valuation: $30 billion (January 2021).

Employees & Offices ​

  • Peak headcount: ~4,000 full-time employees (late 2023)
  • Headquarters: San Francisco, CA
  • Additional offices: Seattle/Bellevue (WA), Pasadena (CA), Phoenix (AZ), Austin (TX), Munich (Germany)
  • Post-restructuring headcount: ~1,050 retained (Feb 2025, after 50% layoff of ~2,100)

Key Milestones Timeline ​

DateMilestone
2013Founded by Kyle Vogt and Dan Kan
Jun 2014RP-1 aftermarket kit announced
Mar 2016Acquired by GM for ~$1B
Oct 2017Acquired Strobe Inc. (solid-state lidar startup)
Oct 2018Honda partnership announced
Nov 2018Dan Ammann appointed CEO
May 2019SoftBank Vision Fund invests $2.25B; valuation $19B
Jan 2020Cruise Origin unveiled (purpose-built AV)
Jan 2021Microsoft invests; valuation hits $30B
Sep 2021CA DMV issues driverless permit
Nov 2021First fully driverless ride in San Francisco
Jun 2022First CA Driverless Deployment Permit (commercial fares)
Feb 20231 million driverless miles reached
Aug 2023CPUC approves 24/7 commercial operations in SF
Oct 2, 2023Pedestrian dragging incident
Oct 24, 2023CA DMV suspends driverless permit
Oct 26, 2023Cruise voluntarily pauses all driverless operations
Nov 2023Kyle Vogt resigns; 24% workforce reduction
Jun 2024Marc Whitten appointed CEO
Dec 2024GM announces end of robotaxi funding; pivot to personal vehicles
Feb 2025GM acquires full ownership; 50% layoff (~1,000 employees); Cruise absorbed into GM

2. Vehicle Platform ​

Chevrolet Bolt EV (Retrofit Platform) ​

Cruise's primary operational vehicle was a modified Chevrolet Bolt EV, retrofitted with autonomous driving hardware. The company iterated through multiple generations:

GenerationApproximate PeriodKey Changes
1st Gen2016--2017Initial Bolt EV integration; early sensor suite
2nd Gen2017--2018Improved sensor placement and compute
3rd Gen2018--2019Unveiled publicly; refined rooftop sensor module
4th Gen2019+Production-intent design; GM filed safety petition with DOT for deployment

Bolt EV Base Specifications (2023 model year):

SpecificationValue
PowertrainAll-electric (single motor, FWD)
Battery65 kWh lithium-ion
Range (EPA)~259 miles
Motor output200 hp / 266 lb-ft
Wheelbase102.4 in
Length163.2 in
Curb weight~3,563 lbs

The Cruise AV modification adds the rooftop sensor module (LiDAR, cameras, radar arrays), additional compute hardware in the trunk area, and removes manual controls in the 4th-gen production-intent variant (no steering wheel, pedals). According to Cruise, 40% of the hardware in the Cruise AV is unique to self-driving and not found in the standard Bolt EV.

Cruise Origin (Purpose-Built, Cancelled) ​

Unveiled in January 2020, the Cruise Origin was a purpose-built autonomous vehicle with no manual driving controls -- no steering wheel, no pedals, no rearview mirrors, and no windshield wipers.

Origin Design Specifications:

SpecificationValue
PlatformGM BEV3 (Ultium architecture)
BatteryGM Ultium (pouch-cell, exact capacity undisclosed)
MotorsUltium Drive
Passenger capacity6 (face-to-face bench seating)
DoorsSliding doors on both sides
LengthApproximately same as Chevrolet Cruze
Entry heightLower and 3x wider than a conventional passenger car
Autonomy levelSAE Level 4--5
ModularityDesigned for sensor/compute upgrades without full fleet replacement
Use casesRide-hail and delivery (convertible interior)

The Origin was designed to be modular, meaning sensor packages and compute units could be upgraded without replacing the entire vehicle. The interior could convert between passenger mode and delivery mode with a slide-in/slide-out delivery unit.

Status: A small number of prototypes were built in late 2023, but no production vehicles were manufactured. The Origin was effectively cancelled when GM ended robotaxi funding in December 2024.

3. Sensor Suite ​

The Cruise AV (Bolt-based) uses a multi-modal sensor suite providing 360-degree coverage:

Sensor TypeCountPurpose
LiDAR53D point cloud mapping, obstacle detection, localization
Cameras16Visual perception, traffic light recognition, lane detection
Radar21Velocity measurement, object detection in adverse weather
Total42Full surround sensing

LiDAR ​

  • 5 LiDAR units mounted on the rooftop sensor bar
  • Early vehicles used Velodyne spinning LiDAR sensors
  • In October 2017, Cruise acquired Strobe Inc., a 12-person startup founded by Lute Maleki (previously of OEwaves), developing chip-scale FMCW (Frequency-Modulated Continuous Wave) solid-state lidar
  • Strobe's technology produces "chirps" of frequency-modulated laser light; measuring the phase and frequency of returning chirps allows simultaneous measurement of both distance and velocity of objects
  • FMCW lidar advantages: relatively immune to interference from other lidar systems, does not require highly sensitive photodetectors, and dramatically reduces cost (target: 99% cost reduction per unit)
  • Cruise pursued a hybrid strategy: using commercial lidar units for near-term deployment while developing proprietary solid-state lidar for future generations

Cameras ​

  • 16 cameras distributed around the vehicle
  • Provide visual perception data for traffic signal recognition, lane markings, signage, and object classification
  • Camera data is fused with LiDAR and radar for robust multi-modal perception

Radar ​

  • 21 radar sensors distributed around the vehicle
  • Provide reliable velocity and range measurements, particularly effective in adverse weather conditions (rain, fog, snow) where lidar and cameras may be degraded
  • Enable detection of objects and their velocity with high precision

Additional Sensors ​

  • Acoustic sensors (microphones) for detecting emergency vehicle sirens and other relevant audio cues
  • GPS/GNSS receivers for coarse positioning (supplemented by LiDAR-based localization in urban canyons)
  • IMU (Inertial Measurement Unit) for dead-reckoning and sensor fusion

4. Onboard Compute ​

Compute Architecture ​

Cruise developed its compute platform significantly in-house, with dedicated hardware engineering teams established from early 2018 under Carl Jenkins (VP Hardware) and Brendan Hermalyn (Director, Autonomous Hardware Systems).

Key architectural properties:

  • Custom AV topology: Encompasses sensors, compute, network systems, connectivity, infotainment, and UX
  • Redundant compute modules: No single point of failure across sensing, compute, networking, or power -- critical because there is no backup human driver
  • High-throughput data processing: The system processes up to 10 gigabits of data per second from the combined sensor suite
  • Real-time processing: Hard real-time constraints for perception, prediction, and planning loops
  • Dedicated silicon architecture: Cruise employed compute and silicon architects to drive system-level architecture decisions, working with leading partners in high-performance computing
  • Hundreds of hardware engineers worked on sensors, network systems, compute, and infotainment systems

Hardware Development Philosophy ​

Rather than relying entirely on off-the-shelf compute solutions (e.g., NVIDIA DRIVE), Cruise invested heavily in custom hardware development and systems integration. This approach provided:

  • Tighter integration between software and hardware
  • Optimization for their specific autonomy workloads
  • Greater control over thermal management, power consumption, and reliability
  • Hardware tailored to the specific sensor suite and processing pipeline

5. Autonomy Software Stack ​

Cruise's autonomy pipeline follows the classical modular architecture with significant machine learning integration:

5.1 Perception ​

  • Sensor fusion: Combines data from LiDAR, cameras, and radar to produce a unified environmental model
  • Object detection and classification: Identifies vehicles, pedestrians, cyclists, traffic signals, signs, and other road objects
  • 3D point cloud processing: LiDAR data processed for obstacle detection and free-space estimation
  • Multi-modal fusion: Camera imagery provides texture and color information fused with LiDAR depth data for robust classification
  • Capable of detecting obstacles even in pitch-black conditions, rain, and fog through complementary sensor modalities

5.2 Prediction ​

  • ML-first approach: Cruise adopted a machine-learning-first strategy for prediction because "people don't necessarily follow the rules of the road"
  • Intent prediction: Predicts future trajectories of pedestrians, vehicles, and cyclists multiple seconds into the future
  • Self-supervised learning framework: Uses future perception output compared against current predictions to create training labels, enabling continuous improvement without manual annotation
  • Continuous Learning Machine (CLM): An automated pipeline for identifying prediction errors, labeling data, training new models, and deploying improvements (detailed in Section 6)

5.3 Planning ​

  • Path planning: Generates safe, comfortable, and efficient trajectories from current position to destination
  • Intersection handling: Uses a purely learning-based approach for complex intersection scenarios
  • Behavioral planning: High-level decision making (lane changes, yielding, merging, unprotected turns)
  • Fallback path planning: Dedicated failover planners for degraded operating conditions
  • Patent filings cover path planners that search for and update optimal plans from a current pose to an end pose while avoiding obstacles

5.4 Control ​

  • Model Predictive Control (MPC): Used for low-level vehicle control (steering, throttle, braking)
  • Kalman Filters: Used for state estimation and object tracking
  • Multiple control modes: Normal operation, minimal risk condition (MRC) handling, and emergency stop capabilities

5.5 Pipeline Integration ​

Sensors --> Perception (Fusion) --> Prediction --> Planning --> Control --> Vehicle Actuation
                  |                      |             |
                  v                      v             v
           HD Map Localization    Scene Context    Safety Monitor

Every piece of code undergoes a physical road test with an engineer in the vehicle before being merged into the production branch, in addition to extensive simulation testing.

6. Machine Learning & AI ​

Continuous Learning Machine (CLM) ​

Cruise's most significant ML innovation is the Continuous Learning Machine, an automated pipeline addressing the "long tail" challenge in autonomous driving:

Three-Step CLM Process:

  1. Error Mining (Active Learning)

    • Automatically identifies scenarios where there is a significant difference between prediction and reality
    • Only problematic scenarios are added to the training dataset, avoiding bloat with "easy" examples
    • Enables extremely targeted data mining

  2. Self-Supervised Labeling

    • Uses future perception output as "ground truth" for prediction scenarios
    • Fully automated -- no human annotators required
    • Enables significant improvements in scale, cost, and speed

  3. Model Training & Evaluation

    • New models are trained, tested through extensive evaluation pipelines, and deployed
    • Metrics pipelines ensure each new model exceeds previous model performance
    • Models must generalize well across diverse scenarios before deployment

Example workflow: If an initial model poorly predicts U-turn situations, the CLM automatically samples U-turn error cases, grows the dataset representation of U-turns, and iterates until the model sufficiently handles them -- all without human intervention.

Training Infrastructure ​

  • Trained on 5+ million miles of real-world driverless driving data
  • Uses Google Cloud Vertex AI to train hundreds of models simultaneously
  • Consumes hundreds of GPU-years every month for model training
  • Models cover perception, prediction, planning, and other AV subsystems

ML Architecture Approaches ​

  • Deep neural networks for object detection and classification
  • Recurrent and transformer-based architectures for trajectory prediction
  • Multi-modal perception models fusing camera, LiDAR, and radar inputs
  • Self-supervised and semi-supervised learning frameworks to reduce labeling dependency
  • Cruise has not publicly disclosed use of a single "foundation model" architecture, but their technology stack includes multimodal perception systems and AI models that are being integrated into GM's next-generation driver-assistance programs

7. Mapping & Localization ​

HD Mapping Approach ​

Cruise produces its own high-definition maps in-house using precision LiDAR and semantic mapping techniques.

Map Contents:

  • Lane boundaries and lane types
  • Traffic light locations and types
  • Curb locations and heights
  • Road surface features
  • Semantic information (crosswalks, stop lines, speed limits, turn restrictions)

Localization Method:

  • Vehicles use LiDAR-to-map matching: the car scans the surrounding environment with lidar and compares it against the HD map to determine its position down to the centimeter
  • This approach is critical in urban environments like San Francisco where tall buildings block GPS signals (urban canyon effect)
  • Map-based localization frees up processing power that would otherwise be spent on environmental understanding, giving the car more compute budget for dynamic maneuvering

Map Maintenance:

  • Cruise developed operational solutions to detect real-world changes (construction, new signage, road modifications)
  • Map updates are pushed to every autonomous vehicle in the fleet within minutes
  • Multiple versions of map features can be A/B tested simultaneously
  • The best-performing version is rapidly deployed fleet-wide

Strategic Advantage:

  • In-house mapping provides full control over maintenance strategy
  • Faster iteration on new map features compared to relying on third-party map providers
  • Ability to rapidly expand to new operational domains

8. Simulation Platform ​

Cruise built one of the most extensive AV simulation platforms in the industry.

Scale ​

MetricValue
Total simulated miles20+ billion
Daily compute jobs200,000 hours of compute per day
Daily instances spun up30,000
Processor cores300,000+
GPUs5,000+
Daily data output~300 TB of results

Simulation Tools ​

ToolFunction
MorpheusRapid generation of specific testing scenarios
Road-to-SimAutomated pipeline that fuses real-world driving data to recreate on-road events in simulation without manual intervention
WorldGenProcedural generation of entire virtual cities for testing in new operational design domains

Scenario Sources ​

Scenarios are drawn from:

  • Millions of miles of real-world driving data collected by the fleet
  • National crash databases (e.g., NHTSA crash data)
  • Academic research on edge cases and failure modes
  • Synthetic generation: entirely new scenarios designed from scratch
  • Modified real-world data: actual events replayed with variations

Simulation Capabilities ​

  • Replay actual on-road events with high fidelity
  • Modify real-world scenarios to create novel edge cases
  • Generate scenarios across different weather conditions, times of day, and traffic densities
  • Test perception, prediction, and planning modules independently or end-to-end
  • Run regression testing against new software releases multiple times per week

Data Warehousing for Simulation ​

Cruise built a dedicated simulation metrics data warehouse using Avro tables. A graph compute engine transforms raw simulation output into structured analytics tables. Engineers can experiment with new simulation metrics without schema migrations, enabling rapid iteration on evaluation criteria.

9. Cloud & Data Infrastructure ​

Multi-Cloud Strategy ​

Cruise operates on a multi-cloud architecture, primarily using two major cloud providers:

Google Cloud Platform (Primary Compute) ​

ServiceUsage
Google Kubernetes Engine (GKE)Primary container orchestration for backend services
Compute EngineVirtual machines for simulation and processing
Cloud StorageLarge-scale data storage for sensor logs and training data
BigQueryAnalytics and data warehousing
Vertex AIML model training (hundreds of models, hundreds of GPU-years/month)
CloudSQLRelational database services
PubSubEvent-driven messaging
Cloud FunctionsServerless compute
App EngineApplication hosting
Cloud Logging & MonitoringObservability

Microsoft Azure (Strategic Partner) ​

  • Following Microsoft's investment in January 2021, Azure became Cruise's primary and preferred cloud provider (though not exclusive)
  • GM uses Azure for collaboration, storage, AI, and machine learning projects
  • The partnership extended to exploring digital supply chain optimization and new mobility services

Data Processing: Terra Platform ​

Cruise developed Terra, a custom data processing platform built as an extension of the Apache Beam SDK:

  • Built in Python on top of Apache Beam
  • Handles dataset registration, lineage tracking, timestamp synchronization, windowing, automatic schema inference, data validation, and feature discovery
  • Provides standard connectors to diverse data stores (raw car data, labeled data, map data, operational data)
  • Weekly usage: 70+ unique users submitting 2,000+ jobs
  • Improved feature engineering pipeline runtime by up to 100x (two orders of magnitude)

Container Platform ​

  • Backend for Cruise self-driving cars runs on Kubernetes
  • Custom tool "Juno" enables application developers to iterate and deploy at scale
  • Multi-tenant, multi-environment Platform as a Service
  • Detailed networking architecture for container platform connectivity

Infrastructure as Code ​

  • Uses HashiCorp Terraform from early days for cloud resource provisioning
  • Library of 150+ versioned, validated, and approved Terraform modules
  • Decomposed mono-repository into well-defined micro-repositories using Terraform Enterprise workspaces
  • Each team has dedicated workspaces for writing and executing infrastructure code

10. Programming Languages & Tools ​

Programming Languages ​

LanguagePrimary Use
C++Real-time AV software (perception, planning, control), latency-critical components
PythonML model training, data processing (Terra), tooling, scripting, simulation analysis
Go (Golang)Backend services, infrastructure tooling, cloud-native applications
Node.jsWeb services, internal tools, dashboards

Build Systems & CI/CD ​

ToolPurpose
CircleCI EnterpriseContinuous integration and deployment
GitHub EnterpriseSource code management and collaboration
Bazel (likely)Build system for multi-language monorepo (common in AV industry for C++/Python)

Infrastructure & DevOps ​

ToolPurpose
DockerContainerization
KubernetesContainer orchestration (via GKE)
HashiCorp TerraformInfrastructure as code
Apache KafkaEvent streaming and message queuing
Apache BeamData processing framework (via Terra)

Internal Tools ​

ToolPurpose
JunoCustom container deployment platform for developers
TerraCustom data processing platform (Apache Beam extension)
MorpheusSimulation scenario generation
Road-to-SimReal-world to simulation conversion pipeline
WorldGenProcedural city/environment generation for simulation

Development Practices ​

  • Monorepo approach (at least historically, per CircleCI case study)
  • Every code change undergoes physical road testing with an engineer in the vehicle before merging to master
  • Simulation-based regression testing enables multiple releases per week
  • New developers are productive from day 1 due to CircleCI/GitHub integration

11. Safety & Redundancy ​

Safety Reports ​

Cruise published formal safety reports (notably the 2022 Safety Report) covering:

  • Operational Design Domain (ODD) definition
  • Regulatory requirements and compliance
  • Safety methodology and approaches
  • High-level architecture and system design
  • Requirements management
  • Verification and validation processes
  • Hardware and firmware verification
  • Cybersecurity validation
  • Test scenario development
  • Safe launch readiness review
  • Operational readiness protocols

Redundancy Architecture ​

The Cruise AV is designed with no single points of failure across four critical domains:

DomainRedundancy Approach
SensingMultiple overlapping sensor modalities (LiDAR, cameras, radar) with overlapping fields of view; system degrades gracefully if one sensor fails
ComputeRedundant compute modules; dual-redundant with active hot standby; safety-critical compute has built-in redundancy
NetworkingRedundant communication lines between computing systems and sensors
PowerDistributed power sources scattered throughout the vehicle (more fail-safe than centralized configurations)

Safety Design Principles ​

  • Fail-operational: System continues operating safely after a single component failure
  • Fail-safe: In the event of multiple failures, the vehicle transitions to a Minimal Risk Condition (MRC) -- typically pulling over and stopping safely
  • Fallback path planning: Dedicated failover planners for degraded operating conditions (covered by patent filings)
  • AI-filtered sensor data: System smoothly transitions between redundant sensors to maintain control if one sensor fails
  • Remote assistance: Fleet operations center can provide remote guidance to vehicles in ambiguous situations

12. Fleet Operations ​

Operational Cities ​

CityLaunch DateOperating HoursCoverage Area
San Francisco, CANov 2021 (driverless)Initially 10pm--6am; expanded to 24/7 (Aug 2023)Initially Sunset/Richmond districts; expanded citywide
Phoenix, AZLate 2022Nighttime initiallySelect neighborhoods
Austin, TXLate 2022Nighttime initiallySelect neighborhoods
Houston, TXOct 20239pm--6am, 7 days/week~11 sq mi (Downtown, Midtown, East Downtown, Montrose, Hyde Park, River Oaks)
Dallas, TXPlanned/limited----

Fleet Size ​

  • Total fleet: approximately 400 vehicles across all cities (pre-pause)
  • Typical new city launch: ~12 vehicles initially, scaling from there
  • All operational vehicles were modified Chevrolet Bolt EVs

Ride Statistics ​

MetricValue
Total driverless miles5+ million (real-world)
1 million driverless miles reachedFebruary 2023 (15 months after first ride)
Total driverless rides delivered250,000+
Driverless miles by Aug 2022250,000+

Pricing (Pre-Pause) ​

ComponentRate
Base fare$5.00
Per mile$0.90
Per minute$0.40
City tax1.5%

Cruise received California's first Driverless Deployment Permit in June 2022, allowing it to charge fares. It became the first company to launch a fared robotaxi service in San Francisco.

13. Regulatory & Safety Record ​

October 2, 2023: Pedestrian Dragging Incident ​

The pivotal incident that precipitated Cruise's downfall:

  1. A pedestrian was struck by a human-driven hit-and-run vehicle, which threw her into the path of a Cruise AV
  2. The Cruise AV, operating driverlessly, ran over the pedestrian
  3. The Cruise AV's software then attempted to pull over to the curb -- dragging the woman approximately 20 feet while she was trapped underneath
  4. The pedestrian suffered serious injuries

Reporting Failures ​

  • Cruise submitted two NHTSA crash reports that omitted the post-crash dragging details
  • NHTSA discovered the omission only after requesting and reviewing video footage from Cruise
  • Cruise initially showed NHTSA and the California DMV a truncated version of the incident video that did not include the dragging portion

Criminal Charges ​

  • Cruise admitted to submitting a false report to influence a federal investigation
  • Paid a $500,000 criminal fine (U.S. Department of Justice, Northern District of California)

Regulatory Actions ​

ActionDateAuthority
CA DMV immediately suspends driverless permitOct 24, 2023California DMV
Cruise voluntarily pauses all driverless operations nationwideOct 26, 2023Self-imposed
24% workforce reduction (~960 employees)Dec 14, 2023Internal
NHTSA consent order with $1.5M penaltySep 2024NHTSA
NHTSA closes preliminary investigationJan 2025NHTSA

Recalls ​

Recall 1 -- Unexpected Braking (2022--2024):

DetailValue
IssueAV software could trigger unexpected hard braking when a cyclist or vehicle approached from the rear
Investigation openedDec 12, 2022
Total hard-braking incidents since 20217,632
Incidents leading to crashes/fires10
Injuries reported4
Vehicles recalled1,194 (entire fleet)
ResolutionSoftware update improving perception, prediction, and path planning; completed during operations pause (Oct 2023 -- May 2024)
Investigation closedAug 2024

Recall 2 -- Post-Crash Behavior (2023):

  • Related to the October 2023 pedestrian incident
  • Software updated to improve post-collision behavior

Other Notable Incidents ​

  • Multiple incidents of Cruise vehicles blocking traffic, including emergency vehicles
  • Vehicles entering wet concrete at construction sites
  • Clustering incidents where multiple Cruise vehicles congregated and blocked streets
  • A Cruise vehicle struck a San Francisco fire truck in August 2023

14. Key Partnerships ​

General Motors (Parent Company) ​

  • Acquired Cruise in March 2016 for ~$1B
  • Provided the Chevrolet Bolt EV platform and manufacturing capabilities
  • Developed the BEV3/Ultium platform for the Origin
  • Completed full ownership acquisition in February 2025
  • Invested cumulative ~$12.1B
  • Now integrating Cruise technology into Super Cruise and future autonomous driving systems

Microsoft ​

  • Invested as part of $2B round in January 2021
  • Azure designated as Cruise's primary and preferred cloud provider
  • Collaboration on software engineering, cloud computing, and AI/ML capabilities
  • Partnership explored digital supply chain optimization

Honda ​

  • Invested $750M equity in Cruise + committed ~$2B over 12 years ($2.75B total commitment)
  • Co-developed the Cruise Origin alongside GM and Cruise
  • Planned joint venture for driverless ridehail service in Japan (targeted early 2026)
  • Joint venture would have used Origin vehicles for 6-passenger autonomous rides
  • Partnership dissolved after GM ended robotaxi development in December 2024

Walmart ​

  • Became a Cruise investor in April 2021 (part of $2.75B round)
  • Piloted self-driving delivery service in Scottsdale, Arizona (November 2021)
  • Partnership explored autonomous last-mile delivery

Strobe Inc. (Acquired) ​

  • Acquired October 2017
  • 12-person solid-state FMCW lidar startup
  • Founded by Lute Maleki (OEwaves spinoff)
  • Technology goal: reduce per-unit lidar cost by 99%

15. Research & Publications ​

Patents ​

GM Cruise Holdings LLC has been assigned approximately 79+ patents covering:

Patent AreaExamples
SimulationPurposeful stress testing of AV response time with simulation
PerceptionSystems responding to adverse weather conditions
Path PlanningSearching and updating optimal plans from current pose to end pose while avoiding obstacles
Camera SystemsCalibration systems for correcting lens distortion
FailoverHandling degraded operating conditions with fallback path planners
Fleet ManagementReal-time autonomous vehicle fleet parking availability

Engineering Blog (Medium) ​

Cruise maintained an active engineering blog at medium.com/cruise with notable publications:

PublicationTopic
"Cruise's Continuous Learning Machine"Self-supervised ML pipeline for prediction improvement
"Introducing Terra"Custom data processing platform built on Apache Beam
"Building a Container Platform at Cruise"Kubernetes-based PaaS architecture
"Container Platform Networking"Kubernetes networking architecture
"Data Warehousing for AV Simulation Analysis"Simulation data infrastructure
"How Cruise Uses Simulation to Speed Up Sensor Development"Sensor-in-the-loop simulation
"3 Ways Cruise HD Maps Give Our Self-Driving Vehicles an Edge"HD mapping approach

Conference Presentations ​

VenueTopic
Google Cloud Next '19"How to Run Millions of Self-Driving Car Simulations on GCP"
MLconf"ML Infrastructure for Autonomous Vehicles @ Cruise"
HashiCorp events"Terraform and Cruise Case Study: A Self-Driven Future"

Academic Engagement ​

While Cruise published fewer academic papers than competitors like Waymo, their engineering contributions focused on:

  • Self-supervised learning for autonomous driving
  • Scalable simulation infrastructure
  • Data processing pipelines for AV applications
  • Container platform engineering at scale

16. Current Status (2025-2026) ​

GM's Strategic Pivot ​

In December 2024, GM CEO Mary Barra announced that GM would stop funding Cruise as a standalone robotaxi business and instead integrate Cruise's autonomous technology into GM's personal vehicle lineup. The rationale:

  • The robotaxi business required too much capital with uncertain timelines to profitability
  • Cruise's technology could be better leveraged for consumer ADAS products
  • GM could reduce spending by $1+ billion annually

Absorption into GM (February 2025) ​

  • GM acquired full ownership of Cruise, completing the merger
  • ~50% of Cruise employees laid off (~1,000 of ~2,100)
  • Departing executives: CEO Marc Whitten, CHRO Nilka Thomas, Chief Safety Officer Steve Kenner, Chief Government Affairs Officer Rob Grant
  • Mo Elshenawy (President/CTO) stayed through April 2025 for transition
  • Retained employees are primarily in engineering roles
  • Sterling Anderson (former Tesla Autopilot chief) hired to lead GM's autonomous driving development

Technology Integration Path ​

Cruise's technology is being channeled into GM's consumer vehicle autonomy roadmap:

SystemTimelineCapability
Super Cruise (current)2018--presentHands-free, eyes-on highway driving; 600,000+ miles of mapped roads; 500,000+ enabled vehicles on road
Enhanced Super Cruise2026 MYGoogle Maps integration; automatic transition between steering assist and hands-free modes
Eyes-Off Driving SystemTarget 2028Hands-off, eyes-off driving starting with Cadillac Escalade IQ; uses lidar + radar + cameras; operates on unmapped highways
Full AutonomyTBDLong-term goal; leveraging Cruise's 5M+ driverless miles and simulation framework

Current Testing (2025-2026) ​

  • GM is using a limited number of Cruise Bolt AVs and other vehicles equipped with lidar on select highways in Michigan, Texas, and the San Francisco Bay Area
  • Vehicles are driven by trained human drivers (not driverless)
  • Testing focuses on developing simulation models and advancing driver-assistance systems
  • Data collection vehicles include Cadillac Escalade IQ and GMC Yukon SUVs gathering driving data from across the U.S.

Cruise Technology Assets Retained by GM ​

  • Multimodal perception systems trained on 5+ million driverless miles
  • AI/ML models and the Continuous Learning Machine pipeline
  • Simulation framework (20+ billion simulated miles, WorldGen, Morpheus, Road-to-Sim)
  • HD mapping technology and rapid map-update infrastructure
  • Sensor fusion expertise and proprietary sensor development (Strobe lidar IP)
  • Cloud infrastructure and data processing pipelines (Terra, Kubernetes platforms)
  • Patent portfolio (79+ patents)

Industry Context ​

Cruise's trajectory from a $30B-valued robotaxi pioneer to absorption into GM's ADAS division represents one of the most significant pivots in the autonomous vehicle industry. While competitors like Waymo continued expanding driverless operations, Cruise's October 2023 incident -- and particularly the cover-up of the pedestrian dragging -- destroyed regulatory trust and public confidence. GM's $12.1B investment, though it did not produce a viable robotaxi business, generated substantial autonomous driving IP that is now being redirected toward the potentially larger market of consumer autonomous vehicles.

Sources ​

Public research notes collected from public sources.

Copyright © 2026 kvynlim