Autonomous Systems Survey: Current State and Future Prospects
Executive Summary
This comprehensive survey examines the current landscape of autonomous systems, with a particular focus on autonomous vehicles, robotics, and the regulatory framework governing these technologies. The analysis covers SAE autonomy levels L3-L5, recent regulatory developments, and provides an in-depth technical overview of leading autonomous vehicle manufacturers in both US and Chinese markets.
Table of Contents
- SAE Autonomy Levels: L3, L4, and L5 Definitions
- Regulatory Landscape and Recent Developments
- US Market: Leading Autonomous Vehicle Technologies
- Chinese Market: Emerging Autonomous Vehicle Leaders
- Technical Analysis and Comparison
- Future Outlook and Challenges
- Conclusion
SAE Autonomy Levels
Level 2 (Partial Automation)
SAE Level 2 represents partial automation where the vehicle can control both steering and acceleration/deceleration simultaneously, but the human driver must remain engaged and monitor the driving environment at all times. Key characteristics include:
- Driver Responsibility: Driver must keep hands on wheel and eyes on road at all times
- Continuous Supervision: Driver is responsible for monitoring the environment and taking immediate control
- Combined Functions: System can handle steering AND acceleration/braking together
- No Legal Disengagement: Driver cannot legally divert attention from driving tasks
- Fallback Responsibility: Driver is the fallback for all driving situations
Examples: Tesla Autopilot/FSD, GM Super Cruise, Ford BlueCruise, Mercedes-Benz Driver Assistance Package
Level 3 (Conditional Automation)
SAE Level 3 represents conditional automation where the vehicle can perform all driving tasks under specific conditions, but the human driver must be ready to take control when requested
- Driver Responsibility: Driver can legally take eyes off the road but must remain alert
- Operational Domain: Limited to specific conditions (highways, traffic jams, mapped areas)
- Speed Limitations: Typically operates under 40-60 mph
- Takeover Requirement: Driver must resume control within seconds when prompted
Key Differences Between Level 2 and Level 3
The transition from Level 2 to Level 3 represents a fundamental shift in responsibility and legal liability:
| Aspect | Level 2 | Level 3 |
|---|---|---|
| Driver Attention | Must monitor environment continuously | Can divert attention under specific conditions |
| Legal Responsibility | Driver responsible for all driving tasks | System responsible during automated mode |
| Hands-on Requirement | Hands on wheel required | Hands-free operation allowed |
| Eyes-on-Road | Eyes must remain on road | Eyes can be diverted (reading, phone use) |
| Fallback Responsibility | Driver is always the fallback | System handles fallback within ODD |
| Regulatory Approval | Minimal regulatory oversight | Requires extensive certification and approval |
Why Most Cars (Including Tesla) Remain at Level 2
Despite advanced capabilities, most manufacturers including Tesla maintain Level 2 classification due to several critical factors:
1. Regulatory and Legal Challenges
- Liability Concerns: Level 3 requires manufacturers to accept legal responsibility for accidents during automated mode
- Certification Requirements: Extensive testing and regulatory approval processes involving multiple organizations:
- SAE International: Defines the J3016 standard for automation levels
- NHTSA (National Highway Traffic Safety Administration): US federal safety standards and approval
- FMVSS (Federal Motor Vehicle Safety Standards): Compliance requirements for vehicle safety
- State DMVs: Individual state permits for testing and deployment (California DMV, Nevada DMV, etc.)
- ISO 26262: Functional safety standard for automotive systems
- UN-ECE WP.29: Global technical regulations for automated driving systems
- TÜV and other certification bodies: Third-party testing and validation
- Example: Mercedes-Benz Drive Pilot required 3+ years of development, extensive real-world testing, and approval from German KBA (Kraftfahrt-Bundesamt) and US NHTSA before certification
- Insurance Implications: Complex insurance frameworks needed for shared liability models
2. Technical Limitations
- Edge Case Handling: Systems struggle with unpredictable scenarios (construction zones, emergency vehicles, unusual weather)
- Sensor Reliability: Current sensor technology has limitations in adverse conditions
- Geographic Constraints: Level 3 systems require detailed mapping and infrastructure support
3. Human-Machine Interface Challenges
- Takeover Time: Ensuring drivers can resume control quickly enough when system reaches its limits
- Attention Management: Difficulty in keeping drivers alert during automated periods
- Trust Calibration: Balancing user confidence without over-reliance on the system
4. Business and Strategic Considerations
- Cost vs. Benefit: Level 3 certification costs may not justify limited operational domains
- Market Positioning: Some manufacturers prefer "advanced Level 2" marketing over restricted Level 3
- Gradual Deployment: Incremental improvement strategy rather than revolutionary jumps
Tesla's Specific Case
- "Full Self-Driving" Branding: Despite the name, Tesla FSD remains Level 2 requiring constant supervision
- Beta Testing Approach: Tesla uses customer fleet for real-world testing rather than controlled certification
- Regulatory Strategy: Avoiding formal Level 3 certification while continuously improving capabilities
- Liability Model: Maintaining driver responsibility reduces legal exposure
Level 4 (High Automation)
Level 4 systems can handle all driving tasks within their operational design domain without human intervention
- Full Autonomy: Within specific geographic areas or conditions
- No Human Intervention: Required during normal operation
- Fallback Capability: System can safely stop if conditions exceed capabilities
- Commercial Applications: Robotaxis, delivery vehicles in defined areas
Level 5 (Full Automation)
Level 5 represents complete autonomy under all conditions that a human driver could handle. No current system has achieved SAE Level 5
Regulatory Landscape
California Public Utilities Commission (CPUC) Approval - August 2023
On August 10, 2023, the California Public Utilities Commission made a landmark decision by approving resolutions granting additional operating authority for both Cruise LLC and Waymo
- 24/7 Operations: Both companies can now operate their robotaxi services around the clock
- Expanded Coverage: Waymo's service extension to parts of San Mateo County
Waymo's Major Expansion (2024-2025)
Service Expansion and Growth:
- Current Operations: Phoenix, San Francisco, and Los Angeles covering over 500 square miles
LA and Bay Area Expansion Details:
- San Francisco: Opened to all residents in June 2024, removing the invitation-only "digital velvet rope"
Safety Improvements:
- Safety performance has improved ~3x since mid-2024
Future Expansion Plans:
- 2025: Austin, Texas and Atlanta (via Uber app), Tokyo testing with Nihon Kotsu
Note on LA Suspension Claims: While there were temporary regulatory discussions in early 2024, no major operational suspension occurred in Los Angeles at that time. However, in June 2025, Waymo temporarily suspended service in downtown Los Angeles after five vehicles were vandalized and set on fire during anti-ICE protests
Other Bay Area Autonomous Vehicle Companies
Zoox (Amazon)
Company Overview:
- Parent Company: Amazon (acquired 2020 for $1.2 billion)
Current Operations (2025):
- Testing Locations: San Francisco Bay Area, Las Vegas, Seattle, Austin, Miami, Atlanta, and Los Angeles
Manufacturing and Production:
- Production Facility: Opened new manufacturing plant in Hayward, California (June 2025)
Regulatory Status:
- NHTSA Exemption: First company to receive demonstration exemption under federal Automated Vehicle Exemption Program (August 2025)
Safety Record:
- Recent Incidents: Software recall of 270 vehicles after Las Vegas collision (April 2025)
WeRide
Company Overview:
- Founded: By Baidu alumni from Tsinghua University
Global Operations:
- Partnership with Uber: Expanding to 15 additional cities worldwide over next five years
Technology:
- Sensor Suite: Over 20 high-performance sensors including high-line LiDARs and high-dynamic cameras
Pony.ai
Company Overview:
- Founded: December 2016 by James Peng and Tiancheng Lou (former Baidu developers)
California Status:
- Permit History: Lost California testing permits in 2022 due to safety driver record violations
Global Operations:
- China Services: Commercial robotaxi operations in Beijing and Guangzhou
Technology and Production:
- Generation 7 Robotaxis: Mass production and road testing launched (July 2025)
Baidu Apollo
Company Overview:
- Parent: Baidu Inc. (China's leading search engine company)
China Operations:
- Wuhan Expansion: Striving to make Wuhan the world's "robotaxi capital"
Market Position:
- Industry Leadership: Key player in China's autonomous vehicle market alongside Pony.ai
US Market Status: - Bay Area Presence: Limited compared to China operations, primarily R&D focused - Regulatory Challenges: No active commercial permits for US operations as of 2025
Cruise LLC Status Update (2023-2024)
However, Cruise's operations faced significant setbacks shortly after this approval:
October 2023 Suspension
On October 24, 2023, the California DMV immediately suspended Cruise's deployment and driverless testing permits
- Safety Incident: A Cruise vehicle struck and dragged a pedestrian approximately 20 feet in San Francisco on October 2, 2023
5 - Regulatory Violations: DMV cited misrepresentation of safety information and determined vehicles were "not safe for public operation"
2 - CPUC Action: California Public Utilities Commission also suspended Cruise's permit for autonomous revenue taxi services the same day
1
Leadership and Organizational Changes
- CEO Resignation: Kyle Vogt resigned as CEO on November 19, 2023
1 - Mass Layoffs: Cruise laid off approximately 25% of its workforce (around 1,000 employees)
1 - New Leadership: GM appointed Mo Elshenawy and Craig Glidden as co-presidents, later replaced by Marc Whitten as CEO
4
2024 Attempted Comeback
- May 2024: Cruise began returning vehicles to public roads with safety drivers for testing
1 - Houston Operations: Resumed limited testing operations in Houston and Dallas with safety drivers
3 - NHTSA Consent Order: Paid $1.5 million fine and submitted corrective action plan for compliance violations
5
December 2024: GM Ends Funding
In a dramatic turn of events, General Motors announced in December 2024 that it would stop funding Cruise entirely
- Strategic Shift: GM will focus on advanced driver assistance systems (ADAS) for personal vehicles rather than autonomous taxis
- Resource Reallocation: Autonomous vehicle technology will be incorporated into GM's Super Cruise system
- Market Reality: Cited "considerable time and resources needed to scale the business" and "increasingly competitive robotaxi market"
- Fleet Status: Approximately 50 Cruise vehicles remain parked in Houston lots as of December 2024
Current Status: Cruise operations are effectively terminated, marking the end of one of the most well-funded autonomous vehicle ventures that had raised over $10 billion in investment since 2016.
Mercedes-Benz Level 3 Certification
In January 2023, Mercedes-Benz became the first company to receive approval for a Level 3 autonomous system in the United States
- Drive Pilot System: World's first certified SAE Level 3 system
- Geographic Availability: California and Nevada (as of late 2023/early 2024)
- Operational Conditions: Under 40 mph on mapped freeways
- Legal Framework: Drivers can legally remove attention from driving tasks
Current Status and Follow-up Developments
Waymo: Continues Level 4 robotaxi operations with significant expansion and safety improvements
Cruise: Faced operational challenges and regulatory scrutiny following the CPUC approval
Honda: Also offers Level 3 capability in Japan, demonstrating international progress
US Market
Tesla Full Self-Driving (FSD) and Robotaxi
Current FSD Status (2024)
- SAE Classification: Level 2 automation requiring continuous driver supervision
1 - Regulatory Status: No state-level permits for autonomous operation as of April 2024
1 - Pricing: $8,000 purchase price or $99/month subscription (as of April 2024)
1
Technical Specifications
- Hardware: 35,000 Nvidia H100 chips for neural network training
1 - Investment: $10 billion cumulative investment by end of 2024
1 - Coverage: Available in US, Canada, China, Mexico, and Puerto Rico
2 - Safety Claims: 54% safer than human drivers when FSD is engaged
2
Key FSD Features
- Navigate on Autopilot (highway)
- Traffic light and stop sign recognition
- City street navigation
- Automatic lane changes
- Summon and parking capabilities
Tesla Robotaxi Launch (2025)
Austin Service Launch:
- Launch Date: June 22, 2025 in Austin, Texas
Scaling and Expansion:
- Initial Fleet: Started with ~10 vehicles, scaling to ~1,000 "within a few months"
Cybercab Development:
- Design: No steering wheel or pedals, fully autonomous design
Regulatory and Market Challenges:
- NHTSA Investigation: Ongoing probe into FSD software after crash reports
BMW i7 and Personal Pilot L3
Level 3 Capability
BMW became the first manufacturer to receive approval for combined Level 2 and Level 3 systems in the same vehicle
- BMW Personal Pilot L3: Available exclusively in Germany for €6,000
- Operating Speed: Up to 60 km/h (37 mph) in traffic jams
- Highway Assistant (L2): Operates up to 130 km/h (81 mph) on highways
- Availability: March 2024 launch in Germany
Technical Features
- Sensor Suite: LiDAR, radar, ultrasonic sensors, cameras
- Night Operation: First L3 system capable of dark conditions
5 - Active Lane Change: Automated overtaking without driver steering input
- Legal Activities: Phone calls, reading, video streaming during L3 mode
Mercedes-Benz Drive Pilot
World's First Certified L3 System
- Certification: First SAE Level 3 system approved in the US
- Geographic Coverage: California and Nevada
- Operating Conditions: Mapped freeways under 40 mph
- Legal Framework: Drivers can legally divert attention from driving
US Market Expansion (2023-2024)
Mercedes-Benz successfully expanded Drive Pilot to the United States, marking a historic achievement as the first automaker to receive US state approval for Level 3 autonomous driving in production vehicles
US Availability Details:
- Launch Timeline: Limited fleet deployment in late 2023, broader customer deliveries in early 2024
Future Expansion Plans:
- Mercedes-Benz aims to expand Drive Pilot to additional US states as regulatory frameworks develop
Volvo EX90 Pilot Assist
Current Capabilities
- SAE Level: Level 2 advanced driver assistance
3 - Future Plans: Level 4 capability planned as paid option
- Safety Focus: Conservative approach prioritizing safety over speed of deployment
- Expected Scope: Highway-focused similar to GM's Super Cruise
Chinese Market
XPeng XNGP (XPeng Navigation Guided Pilot)
System Capabilities
- Coverage: Nationwide rollout completed by end of 2024
4 - City NGP: Advanced urban driving in Shanghai and Guangzhou
- Technical Features: Traffic light recognition, lane changes, overtaking, turns
- Map Independence: Operates without high-precision mapping
4
Recent Developments (2024)
- AI Tianji System: XOS 5.2.0 with 484 upgraded features
4 - AI Hawkeye Visual Solution: LiDAR-equivalent performance using cameras only
- Global Expansion: Plans to bring self-driving tech worldwide in 2025
- Volkswagen Partnership: $700 million investment for joint development
Li Auto NOA (Navigate on Autopilot)
Market Position
- Sales Leadership: 500,000 second-generation cars sold, 50,035 in December 2024
1 - Free System: No additional charge for NOA on L9 and Mega models
- Data Advantage: Largest data collection due to standard inclusion
Technical Architecture (2024)
- End-to-End Model: Complete vehicle control system
- Vision-Language Model: Large-scale VLM integration
- Map-Free NOA: Operation without detailed mapping
- Next-Gen Chips: Nvidia Drive Thor for 2025 models (2,000 TOPS performance)
NIO NAD (Navigate on Autopilot)
System Specifications
- Hardware: All second-generation cars equipped with necessary sensors
- Pricing Model: $530/month subscription fee
1 - Custom Chip: Shenji NX9031 (5nm, 50+ billion transistors) for ET9 flagship
- Performance: Comparable to four Nvidia Drive Orin X chips
Xiaomi SU7 Autonomous Features
Vehicle Overview
- Launch: March 2024 with immediate market success
- Performance: SU7 Max: 0-100 km/h in 2.78 seconds
2 - Ultra Variant: 1,139 kW (1,527 hp) with track-focused performance
- Nürburgring Record: Sub-7-minute lap time, faster than Tesla Model S Plaid
Autonomous Technology
- Sensor Suite: LiDAR module available on higher trims
- Drag Coefficient: World's lowest at 0.195 (without LiDAR)
- Integration: Seamless connection with Xiaomi ecosystem
- Market Impact: Strong consumer confidence due to Xiaomi's brand recognition
Technical Analysis
Sensor Technologies
LiDAR vs Camera-Based Systems
- Tesla Approach: Pure vision system with neural networks
- Chinese Manufacturers: Hybrid LiDAR + camera systems
- BMW/Mercedes: Multi-sensor fusion including LiDAR, radar, cameras
- Cost Considerations: LiDAR adds significant expense but improves reliability
Processing Power Evolution
- Tesla: 35,000 Nvidia H100 chips for training
- Li Auto: Nvidia Drive Thor (2,000 TOPS) for 2025
- NIO: Custom Shenji NX9031 chip (50+ billion transistors)
- Industry Trend: Move toward custom silicon for competitive advantage
Software Architectures
End-to-End Learning
- Li Auto: Complete E2E model implementation
- Tesla: Neural network-based approach
- Traditional OEMs: Rule-based systems with ML components
Map Dependency
- High-Definition Maps: Mercedes, BMW rely on precise mapping
- Map-Free Systems: XPeng, Li Auto developing map-independent solutions
- Hybrid Approaches: Combination of HD maps and real-time perception
Performance Comparison
| Manufacturer | SAE Level | Speed Limit | Geographic Coverage | Pricing Model |
|---|---|---|---|---|
| Tesla FSD | L2 | No limit | US, Canada, China, Mexico | $8,000 or $99/month |
| BMW Personal Pilot L3 | L3 | 60 km/h | Germany only | €6,000 |
| Mercedes Drive Pilot | L3 | 40 mph | California, Nevada | Included in S-Class |
| XPeng XNGP | L2+ | Variable | China nationwide | Included in Max trims |
| Li Auto NOA | L2+ | Variable | China | Free on L9/Mega |
| NIO NAD | L2+ | Variable | China | $530/month |
Future Outlook
Technology Trends
Artificial Intelligence Integration
- Large Language Models: Integration for natural interaction
- Computer Vision: Advanced object recognition and prediction
- Edge Computing: Real-time processing in vehicles
- 5G Connectivity: Enhanced vehicle-to-everything (V2X) communication
Hardware Evolution
- Custom Silicon: Manufacturer-specific chips for competitive advantage
- Sensor Fusion: Improved integration of multiple sensor types
- Quantum Computing: Potential future application for complex optimization
Market Consolidation
Industry analysts predict significant consolidation in the Chinese EV market, with Professor Zhu Xican stating that the probability of NIO, XPeng, and Li Auto surviving independently is "zero"
- Scale Requirements: 2 million vehicles annually needed for survival
- R&D Costs: High investment requirements for competitive technology
- Market Dominance: BYD's cost control and Tesla's scale advantages
- Consolidation Pressure: Need for mergers and strategic partnerships
Regulatory Evolution
Global Harmonization
- International Standards: Efforts to align SAE and ISO standards
- Cross-Border Recognition: Mutual recognition of certifications
- Liability Frameworks: Clear legal responsibility for autonomous systems
Safety Validation
- Testing Requirements: Standardized validation procedures
- Real-World Data: Emphasis on actual performance metrics
- Continuous Monitoring: Post-deployment safety assessment
Challenges and Limitations
Technical Challenges
Edge Cases
- Unpredictable Scenarios: Construction zones, emergency vehicles
- Weather Conditions: Snow, heavy rain, fog impact sensor performance
- Infrastructure Variability: Inconsistent road markings and signage
Human-Machine Interaction
- Takeover Requests: Ensuring driver readiness for control resumption
- Trust Calibration: Appropriate reliance on automated systems
- Interface Design: Clear communication of system status and limitations
Regulatory Hurdles
Liability and Insurance
- Accident Responsibility: Determining fault in autonomous vehicle incidents
- Insurance Models: New frameworks for coverage and claims
- Legal Precedents: Establishing case law for autonomous systems
Public Acceptance
- Safety Perception: Building consumer confidence in autonomous technology
- Job Displacement: Addressing concerns about employment impact
- Privacy Issues: Data collection and usage transparency
Conclusion
The autonomous vehicle industry stands at a critical juncture, with Level 3 systems beginning commercial deployment while Level 4 and 5 capabilities remain largely experimental. The regulatory approval of Mercedes-Benz's Drive Pilot and BMW's Personal Pilot L3 systems marks a significant milestone, while the CPUC's approval of Waymo and Cruise robotaxis demonstrates the potential for Level 4 commercial applications.
Chinese manufacturers, particularly XPeng, Li Auto, and NIO, are rapidly advancing with comprehensive autonomous driving solutions, often surpassing traditional automakers in software capabilities and deployment speed. However, market consolidation pressures and the need for massive scale suggest that only a few players will survive independently.
The path to full autonomy remains challenging, with technical hurdles around edge cases, regulatory frameworks still evolving, and public acceptance requiring continued demonstration of safety and reliability. Success will likely depend on a combination of technological innovation, regulatory cooperation, and strategic market positioning.
As the industry moves forward, the integration of AI, advancement in sensor technologies, and development of robust safety validation frameworks will be crucial for realizing the full potential of autonomous systems across transportation, robotics, and other applications.
This survey represents the current state of autonomous systems as of 2024, with particular focus on developments through late 2024 and early 2025. The rapidly evolving nature of this field necessitates regular updates to maintain accuracy and relevance.