Back in 2009, Waymo's "Caddy" served as a Pioneering Campus Autonomous Taxi, predating the concept of Cybercabs.
Recently, Tesla staged its highly anticipated "We Robot" event to present its concept robotaxi, offering rides to attendees on a secluded course at the Warner Brothers' studio lot. Interestingly enough, around fifteen years prior, in 2009, Google Chauffeur, now known as Waymo, conducted a similar demonstration. Utilizing budget-friendly golf carts and bearing the nickname "Caddy," this primitive version of Waymo's autonomous vehicles was a daring endeavor with limited resources.
Caddy was essentially the preliminary version of Waymo's self-driving vehicles. I happened to joining the Chauffeur team as the Caddy project was wrapping up. The initiative was a low-cost experiment, operated by a single full-time engineer, along with part-time support from other members of the 15-person team. The team equipped golf carts with entry-level cameras, ultrasonics, a wheel encoder, IMU/GPS, and 2009-era computers. Advanced technologies like Deep Learning, Transformers, and LIDARs, which would become commonplace in the future, did not exist yet, and were deliberately avoided to explore the possibilities without them.
You can watch a video of Caddy in motion here:
Caddy boasted a collection of pick-up points scattered around the Google campus. Users could summon a vehicle to their stop using a web-based platform or an app, which was still in its infancy at the time. The cart could drive guests autonomously on a restricted range of non-public routes. Users could also choose to bypass the route map and steer the cart on public roads to reach destinations within the campus. In order to cross public roads, the carts could communicate with intersection boxes equipped with radars and cameras to verify that no other vehicles were approaching. In those days, even radars were considered expensive compared to the cost of golf carts.
Caddy served as a proof-of-concept project to assess the capabilities of low-cost hardware. The stereo vision cameras were even scavenged from old PlayStation consoles. At the time, the more expensive Toyota Prius was also being developed by the main team. Caddy was scarcely used, and never saw commercial production. It did perform autonomous operations on a few restricted routes, primarily on sidewalks and golf cart paths, although it did so without the need for a human supervisor or "safety driver." Despite its limited capabilities, it paled in comparison to the self-driving systems used during Tesla's "We Robot" demo held recently.
Remarkably, Caddy demonstrated a working robotaxi service on fixed routes within a mostly enclosed campus, which could be summoned and controlled via a mobile phone app. This capability was not remarkably different from the level demonstrated by Tesla or other similar demos that have surfaced over the years. Regardless, Caddy's team was composed of talented engineers, such as lead Nathaniel Fairfield and perception engineer Jiajun Zhu (now CEO of Nuro). The true impact of Caddy, however, lies in the understanding that even after such demos, significant advancements are still required. Fifteen years later, Waymo has progressed to its sixth-generation vehicle, expanding operations across multiple cities, and introducing self-driving services to the general public in three cities with three more in the pipeline.
Infrastructure sensors, like the intersection radars employed by Caddy, remain a topic of debate in the industry to this day. While they offer a cheaper alternative and can detect things cars cannot, they also restrict vehicles to only the locations where sensors have been installed. This reliance on infrastructure poses challenges due to the slow pace of infrastructure development and the tendency to become obsolete before completion. Nonetheless, companies like Baidu are pursuing solutions like this in China, where infrastructure is typically newer and is developed more rapidly. Mostly, however, robocar teams have opted to follow the principle of "driving on the roads given, trying not to change the world to suit the vehicle." I've even coined this concept the "First Law of Robocars." Nevertheless, there are projects, such as Cavnue in Michigan, striving to make roadside infrastructure a reality.
Waymo's first generation of autonomous vehicles was based on the Prius, utilizing the high-end Velodyne LIDAR, as well as numerous other sensors. The LIDAR provided the majority of the processing power, while the camera was mainly used to detect the color of traffic lights. The vehicle proved capable of navigating 1,000 miles of various road types and accumulating 100,000 miles of operation to meet Larry Page's initial milestone, despite frequent interventions from safety drivers. Today, Waymo's fifth-generation vehicles are offering 150,000 rides per week to members of the general public, with no human intervention required, a fortunate occurrence as incidents are rare.
After Tesla's "We Robot" event, some compared its self-driving technology to Google's early efforts with its 'Caddy' project. Driven by a single engineer in 2009, 'Caddy' was an autonomous golf cart created by Google's Chauffeur team, precursor to Waymo's self-driving vehicles. Cybercabs, like Tesla's concept robotaxi, are now pursuing the same goal, leveraging advancements in technology that weren't available during Caddy's development, such as Deep Learning, Transformers, and LIDARs.
