TuSimple

Type

• Neural networks applied to computer vision
• Asserting AI is based on convolutional neural network technology.

  AI use

• Computer vision (object detection) – advancing technology
  • See two lead scientists’ publications:
    • http://www.houxiaodi.com/publication.html
    • http://winsty.net/publication.html
• Support service providers
  • AI drives trucks that transport goods – enables trucks to drive longer than humans can, decreasing number of hours for long-haul trips (e.g., 5-day trip can be done in 2 days)
  • AI tracks, coordinates and finds efficiencies for customer fleets – reduces time, labor costs, risks of labor shortage, gas used and wear on breaks (AI prediction optimizes use of throttle and brakes less)
• Enable resource integration between service providers and beneficiaries
  • Assumed to coordinate network more efficiently given high-degree of automation – not specifically stated
• Support beneficiaries’ well-being (e.g., end-customers)
  • Goods travel long distances in shorter time at reduced cost
  • Roads are safer for drivers – risk of driver fatigue eliminated and AI decision making response time 15x faster than human’s

AI useRate of return on customer’s investment to make AI work

Immediate:

• OEMs and component suppliers – Reduced time to access specialized IP required to design new trucks to meet AV needs
• Shippers and Fleet owners – Increased awareness of what new technology will mean for industry
• Shippers and Fleet owners – small increase in capacity

Long term:

• Shippers and Fleet owners – addresses significant risk in capacity due to growing labor shortage in drivers
• Shippers and Fleet owners – increased profitability by decreasing cost of labor (40% of revenue in US) and other variable costs (optimized driving)
• Shippers and Fleet owners – decreases time to arrival for goods that require shipping distances of over 8 hours of driving
• Truck drivers – less time out of town, less risk of driver’s fatigue and accidents

Databases

• Initially used data acquired from mounting cameras on existing trucking companies’ trucks
• Continuously increases proprietary database of data captured on their own trucks

Quantum Computing

• N/A

Assets

• Proprietary computer vision perception algorithms, driving algorithms, and traffic prediction algorithms
• Purpose built technology co-created with Tier 1 suppliers
• Proprietary designs for truck components specifically made for AVs
• Partnerships with large tech and service incumbents such as Nvidia, UPS, US Postal Service, Peterbilt, Mando Corp, Penske, US Xpress, etc., which increase their access to advanced technologies, new tech developed with their specifications, complementary capabilities and future customers
• Strategic cooperation agreement with the Caofeidian District of Tangshan in the Hebei Province of northeastern China (near Beijing) that allowed TuSimple to test autonomous trucks on public roads and create a zone to commercially operate an automated logistics zone
• Regulatory permission from US state and federal governments
• Highly technical founding team and research partners (university research centers)

Processes

• Train driving decision-making algorithm using deep learning on a combination of multiple computer vision sensor inputs (LiDAR, radar, and high definition cameras)
• Retrofit trucks to be autonomous
• Partnership acquisition
• Internal coordination to manage co-creation projects with world leading technology companies
• Fundraising
• Government regulatory liaising
• 3D mapping of shipping routes

Priorities

• Build a camera-centric perception solution as the core technology instead of LiDAR centric
  • Use deep learning super-computer plus low-cost vision and radar sensors to make commercialization
  • easier, and then integrate LiDAR at a later stage
• Build team with deep technology expertise from both China and the USA by establishing R&D
  facilities close to talent (in Beijing and San Diego) (60% of early team were PhDs)
• Train computer vision enabled deep learning with data collected aboard manually driven trucks
• Focus on a production level system: find the fastest path to commercial application
• Focus on mastering highway driving scenarios for faster commercialization without human
  intervention
• Build out fleets in both US and China by first retrofitting trucks to be autonomous and then have
  partner companies build the trucks they require to operate mature business model (one brain, many
  trucks)
• Raise money between each major milestone demonstrated to market
• Acquire strategic investors that are upstream and downstream in the value chain
• Partner with original equipment manufacturers (OEMs), Tier 1 suppliers, and sensor technology
  companies to access production-level quality components that are designed to integrate into the
  autonomous system solution (e.g., powertrain, braking and steering systems)
• Take an ecosystem approach by partnering with shippers, technology companies and service providers to
  expand service coverage
  • Phase 1: Two cities in Arizona and four cities in Texas
  • Phase 2: Integrate LA and all major cities in the South of continental USA (South of Nashville)
  • Phase 3: Rest of USA
• Complete an IPO to raise at least a USD$ 1 billion to fuel expansion of US AFN