A Baidu Apollo Park visit in Beijing takes a delegation inside the research and testing base that Baidu built in Yizhuang, the economic and technological development area in the south-east of the capital. The site covers 17,000 square metres of engineering workspace spread across laboratories, vehicle preparation areas and a traffic-control centre, with more than 300 autonomous test vehicles attached to it and 750 kilometres of two-way open test road within reach of the gate. The programme is built around how the technology is developed and proved rather than how it looks at a product launch: the perception, prediction, planning and control stack, the simulation and data loop behind it, the roadside sensors and smart intersections that make up China's vehicle-road-cloud approach, and the safety-driver and remote-assistance layer that still sits behind a driverless claim. Teams leave with a working picture of the engineering constraints and of the operating arithmetic of a paid robotaxi fleet.

Apollo is Baidu’s autonomous-driving platform, while Apollo Go is its autonomous ride-hailing service brand. Through the exhibition and passenger experience, participants can connect the development of autonomous driving with products, services and everyday mobility.

China Innovation Visits coordinates company reception and the exchange according to the delegation’s background, learning objectives and language requirements.

Key takeaways

Apollo Park is Baidu's autonomous-driving research and testing base in Yizhuang, south-east Beijing — 17,000 square metres of engineering space, with more than 300 test vehicles and 750 kilometres of two-way open test road on its doorstep.

It is a development campus, not an exhibition: engineering teams, a vehicle yard, a traffic-control centre and a test regime that escalates from simulation to closed course to public road. What you walk through is the process standing behind the product.

The numbers that matter commercially are public. The sixth-generation robotaxi is quoted at about 200,000 yuan against 480,000 yuan for the first generation, at 0.5 yuan per kilometre and roughly 25 orders per vehicle per day. The paid service reported over 213,000 orders in late 2021 and reached 30 cities by 2025.

The least discussed layer is the one worth understanding best: safety drivers and remote assistance. Driverless passenger demonstration was permitted across 60 square kilometres of Yizhuang from 2022, yet supervision from outside the vehicle remains part of the operation, and the ratio of supervisors to vehicles is a fair question for the host. We arrange the programme end to end, including host introduction, scheduling and interpretation on site.

Location and travel

Item Detail
Location Yizhuang (Beijing Economic and Technological Development Area), southeastern Beijing
Format Exhibition tour, autonomous ride and thematic sharing and exchange
Duration Approximately two and a half hours; hotel transfers are additional
Group size More than 10 and fewer than 20 participants; smaller groups may apply
Lead time Planning at least 30 days ahead is recommended
Photography Prior permission required; restrictions apply in certain areas
Interpretation English or other language support arranged on request

The programme takes place in Yizhuang, the Beijing Economic and Technological Development Area in southeastern Beijing. The district brings together intelligent connected vehicles, robotics and advanced manufacturing, allowing the visit to connect with other technology programmes.

Transfers can be coordinated from the delegation’s hotel or arrival point. The meeting location, visitor entrance and travel time are confirmed during planning according to company reception arrangements and traffic conditions.

Location — Apollo Park sits inside the Yizhuang Economic and Technological Development Area in south-east Beijing, in the district designated as the core of the city's high-level autonomous driving demonstration zone. For planning, the district matters more than a street number: Yizhuang is a purpose-built industrial and technology area rather than a central neighbourhood.

Getting there — Yizhuang is roughly 45 to 60 minutes by road from central Beijing, longer in the morning peak. Delegations normally stay in the city and travel out for the day. Transport is part of the programme, including hotel pickup, the return leg and a briefing in the vehicle.

Duration — Half a day for the guided technical walk and the discussion session, or a full day when the visit is combined with another Yizhuang site or extended with a session on strategy and regulation.

Format — A guided walk through the working areas the host makes available, a briefing on development and testing, a look at the traffic-control and remote-assistance systems where available, and a discussion with company representatives. Test areas can be unavailable at short notice when vehicles or engineers are committed elsewhere.

Best suited to — Automotive and component teams, transport and city-planning groups, AI and systems engineers, infrastructure and telecom suppliers, investors, and regulators studying how driverless services are permitted, monitored and supervised.

Main themes — Sensor fusion and perception, prediction and planning, simulation and data closed loops, vehicle-road-cloud integration, open-road testing and permits, safety-driver and remote-assistance requirements, and the unit economics of a paid robotaxi service.

Why visit Baidu Apollo Park in Beijing?

The exhibition provides an overview, while the ride offers direct observations. Travelling in an autonomous vehicle allows participants to consider its driving behaviour, turns and stops from a passenger’s perspective.

The subsequent exchange connects those observations with business management, product innovation and practical applications. Automotive, transport, logistics and other business teams can gain a reference for understanding AI adoption and service development.

Most technology sites a delegation can visit are polished: the product works, the process is settled and the tour explains a finished thing. Apollo Park is the opposite kind of visit. It shows a technology still being proved, and the useful material is not the demonstration itself but the failure cases and the cost decisions behind it.

That matters to teams facing the same class of problem elsewhere: a system that must work in the messy physical world, where the last few per cent of accuracy decides whether it can operate commercially. Autonomous driving is the most expensive open example of that problem, and the habits visible here transfer far beyond vehicles.

What the visit includes

The park is arranged as an engineering sequence rather than a showroom route: labs where the software is written, a vehicle yard where the test fleet is prepared, a control centre where traffic data and remote supervision meet, and test environments that escalate from simulation to closed course to open road. Each stage catches a class of failure before the next stage sees it.

Exhibition tour

Visit the Baidu Apollo exhibition to learn about the development of autonomous driving, representative vehicles and smart-mobility applications.

The available displays introduce how vehicle sensing, road information and supporting systems work together. Participants can also learn about the Apollo Go service model and its operating functions. Exhibition content and the visitor route follow the company’s arrangements.

The software is a classic autonomy stack: perception turns camera, lidar and radar returns into a model of the world, prediction estimates what people and vehicles will do next, planning chooses a trajectory and control executes it. Hosts describe the difficulty in everyday terms — a roadside stall at a hutong corner, a manhole submerged after heavy rain, a rider appearing between parked cars. Reliability then shows up in the specification: an ASIL-D level-4 stack with dual controllers and dual braking carries redundancy through the vehicle, and redundancy does not vanish because the software improves.

Autonomous ride

Experience an autonomous vehicle under the arrangements and guidance of the on-site team.

Participants can observe how the vehicle responds to traffic signals, surrounding vehicles and road conditions, while considering ride comfort and the in-vehicle interface. Vehicles, routes and boarding groups are arranged according to operating conditions.

Every test vehicle is also a data-collection platform. The loop runs like this: vehicles drive, on-board filtering keeps what is worth keeping, unusual sequences are mined from the fleet, labellers annotate them, the model is retrained, and each new version is re-validated in simulation before returning to the road; Baidu reported road-test data passing 100 million kilometres in 2025. This is why the loop, not the algorithm, is the scarce asset: architectures are published and engineers move between companies, while a permitted fleet and an instrumented city do not. Data also carries obligations, including transfer rules worth asking about.

Thematic sharing and business exchange

The sharing session is prepared around the customer’s needs. Relevant subjects include Baidu’s autonomous-driving development, Apollo Go operations, smart mobility and AI applications.

The theme and content can be customised to the delegation’s priorities, with time for questions and discussion. Speaker arrangements and the scope of the session are coordinated with the company during planning.

Alongside the vehicle sits an infrastructure layer. The park's traffic-control centre, built around Baidu's ACE engine, is described as connecting smart signals and roadside sensors across the city, with congestion on a major avenue forecast about 30 seconds ahead. The related smart-intersection solution covers roughly 12.1 kilometres in Yizhuang, with reported throughput gains near 30 per cent, and the wider programme cites more than 1,000 intelligent roadside units exchanging vehicle-to-road information in under 10 milliseconds. Gantry-mounted sensors see around corners and past the lorry blocking your view. The constraint is that roadside capability is public infrastructure, funded and maintained by someone other than the operator, and it exists only where installed.

Understanding the operations behind the ride

An autonomous passenger journey depends on cooperation between vehicles, operating teams and supporting systems. Dispatch, vehicle monitoring, remote assistance and passenger service contribute to sustained operations.

Remote assistance is the layer marketing compresses. A driverless service runs in stages: a safety driver in the vehicle while a route is being proved, then supervision from outside, where operators watch a fleet and can intervene. The park demonstrates the mechanism in 5G remote-driving cabins linked to a test vehicle about 20 kilometres away, and hosts describe cloud-side takeover in heavy rain. The honest accounting is about ratios and latency: supervision does not disappear when the driver leaves the car, it moves to a control room, and how many vehicles one operator can safely monitor sets a ceiling on how cheap the service can become. The rules on when supervision is required, and at what ratio, are set by the authorities and have been revised in phases.

Autonomous driving and smart transport

Autonomous driving develops within a wider environment of roads, transport infrastructure and city management. Vehicle capability must work with the conditions of the area where the service operates.

Testing on open roads in Yizhuang

The test fleet has access to 750 kilometres of two-way open test road, and Yizhuang works as a graduated proving ground: closed course, then designated public roads, then paid passenger service. The commercial step came in 2022, with one of Beijing's first notices for driverless passenger demonstration covering 60 square kilometres of the district. Open-road testing means real pedestrians, scooters, roadworks and weather, but a demonstration zone is still curated: meticulously mapped, partly instrumented, and populated by drivers who mostly know the vehicles are there. The question worth putting to the host is how much of that tidiness is designed in, and what changes when the same software runs in an unmapped city.

What Apollo offers other carmakers

Apollo is also a supply proposition. Instead of every manufacturer building its own autonomy stack, the platform model offers driving software, maps and validation tooling as something a carmaker can adopt; the ecosystem around the park shows the strategy, with mapping and connected-vehicle suppliers, established and newer vehicle brands, and local industrial funds. The same platform runs outside passenger cars, including unmanned heavy trucks with a logistics partner at a reported monthly freight volume above 1.2 million tonnes. The tension is ownership: a carmaker adopting someone else's stack gains speed and loses differentiation, while the platform owner gains volume and a share of a relationship it did not previously have. Baidu's wider AI stack, including its PaddlePaddle deep-learning platform, sits underneath this proposition.

Autonomy can be built in two structurally different ways, and Yizhuang shows both. The first concentrates intelligence in the vehicle: enough sensing, compute and software that each car can decide on its own. The second distributes part of the sensing and decision-making into the road — smart signals, roadside units, edge computing and a network link — so the vehicle receives information it could not have gathered itself.

The trade-offs run in opposite directions. On-vehicle intelligence puts the full cost into every unit sold and must work wherever the car goes. Roadside intelligence moves much of that cost into shared infrastructure and can see around occlusions, but it works only where coverage exists and stops at the boundary of the equipped area.

Which architecture a market settles on is therefore an institutional question as much as a technical one: who funds the roadside layer, how quickly interface standards are agreed, and what regulators permit. For a delegation, the value of this site is that both halves — the vehicle stack in the labs and the infrastructure layer outside — can be examined in one day.

Who the programme suits

The programme is suitable for business leaders, automotive and transport teams, AI and technology companies, logistics and operations managers, industry associations, and university or business-school groups.

A technical background is not required. The emphasis of the sharing session can be coordinated according to the participants’ industry background and learning objectives.

What different teams take away

The same walk produces different notes depending on who holds the pen. Decide in advance which of these your delegation is, because that determines which questions are worth the host's time.

Your interest is the boundary between software and hardware. Ask which sensing and computing functions the platform expects the vehicle to provide, how redundancy is specified and tested, and how a carmaker integrates an external stack without losing control of the customer experience.

Your interest is the roadside layer and the rules around it. Ask who funds smart intersections, what maintenance costs after installation, how roadside data is governed, and what a city gains beyond autonomous vehicles — the reported junction throughput numbers matter most here.

Your interest is the loop: data selection, labelling quality, simulation fidelity, and how a new model version is gated before release. Ask what proportion of issues is found in simulation rather than on the road, and how the test set is refreshed as the fleet learns.

Your interest is the permission structure and the unit economics. Ask what demonstration permits allow by phase, what reporting the operator owes the authorities, how incidents are handled, and which cost lines dominate per-kilometre economics once the vehicle is genuinely driverless.

How to arrange a Baidu Apollo Park visit in Beijing

China Innovation Visits coordinates visits in advance, with reception confirmed according to the company’s schedule. Planning at least 30 days ahead is recommended.

Provide the preferred date, group size, participant background, learning objectives and language requirements. These help coordinate boarding groups, the thematic exchange and interpretation.

The programme can be customised to the delegation’s needs. The visitor route and reception arrangements are confirmed before the visit.

Visit programme

09:30–10:30 — Exhibition tour and autonomous ride

Visit the Baidu Apollo exhibition to learn about autonomous-driving development and smart-mobility applications, then experience an autonomous vehicle according to the on-site arrangements.

10:30–12:00 — Thematic sharing, business exchange and Q&A

Participate in a session prepared around the customer’s needs. The content can be customised, with discussion addressing the delegation’s priorities.

The programme lasts approximately two and a half hours. Hotel transfers are additional.

Specific arrangements, timings and the itinerary are subject to the final actual programme.

Combine with other company visits

The programme can form part of a Beijing technology tour, with additional companies selected according to the delegation’s learning objectives.

Teams interested in electric vehicles and manufacturing can combine it with a Xiaomi EV factory tour, connecting mobility applications with automotive production. Groups interested in robotics and embodied AI can consider Robot World in Beijing.

The programme runs at the park in Yizhuang, south-east Beijing, and the district is part of the visit: the test roads, the instrumented junctions and the surrounding supplier ecosystem are all a short drive away. Because the area holds a cluster of automation and robotics businesses, the day can be extended with a second stop such as Robot World in Beijing.

It is worth being precise about what this site is, because there is more than one Baidu Apollo Park. Shanghai's park is the operating and demonstration base in Jiading; this one is the research and testing site inside Beijing's Yizhuang autonomous-driving cluster. Teams wanting the commercial side of the service should read our notes on the Baidu Apollo Park in Shanghai.

For longer itineraries, Beijing combines well with other Chinese technology hubs, and autonomy content sits naturally beside robotics, battery and semiconductor stops. We build these across cities so the same questions can be put to more than one operator; an overview of what a company visit programme can include is the place to start.

Practical notes

Comfortable casual shoes are recommended for walking and standing during the visit. Business-casual clothing is suitable.

Follow staff instructions during the autonomous ride. Photography, filming and recording require advance confirmation and must follow the company’s on-site requirements.

Closed-toe shoes are required and sandals will not be permitted in the working areas. Comfortable shoes matter: the programme involves standing and walking through engineering spaces rather than sitting in a conference room.

Photography is restricted. Assume phones and cameras are not permitted on the technical route unless the host has confirmed otherwise, and expect permitted photography to be limited to designated areas. Test areas can also close at short notice; the mitigation is a flexible itinerary and an equivalent substitute rather than a cancelled half-day.

English interpretation is provided on site. The vocabulary of autonomy — perception, prediction, planning, lidar, safety case, remote assistance — does not always translate cleanly into Chinese engineering usage, so we brief the interpreter on your delegation beforehand and ask for a short list of the terms that matter most.

The value of the day is not the vehicle demonstration. It is watching a company account for the last one per cent of reliability, and seeing what that last per cent costs.

Questions before a Baidu Apollo Park visit

What group size can be accommodated? Visits are normally arranged for groups of more than 10 and fewer than 20 participants. Smaller groups may apply for reception, subject to confirmation by the company.

Can international business delegations visit? International delegations may apply. Company reception and language support require advance coordination and confirmation.

Is the autonomous ride included? Yes. The standard programme includes the exhibition tour, autonomous ride and thematic exchange. Boarding groups and routes follow the on-site arrangements.

Can we take photographs? Prior permission is required. Photography may be restricted in certain areas or for particular displays.

What happens if the ride arrangements change? If weather, road conditions or operations affect the ride, its timing and related arrangements are adjusted in coordination with the company.

How long does the visit take? The standard programme takes approximately two and a half hours, from 09:30 to 12:00. It extends to a full day when the visit is combined with other stops or with an additional session.

Is the visit suitable for people from outside the automotive industry? Yes. The programme also suits business leaders, logistics and operations managers, AI and technology companies, industry associations, and university or business-school groups. A technical background is not required.

Can the visit be combined with other companies? Yes. The programme can form part of a Beijing technology tour, with additional companies selected according to the delegation’s learning objectives, including a Xiaomi EV factory tour for teams interested in manufacturing.

What preparation do participants need? No technical background is required. Closed-toe shoes are required and sandals are not permitted in the working areas, and comfortable shoes matter because the programme involves standing and walking. We send a short briefing note beforehand.

Arrange your Baidu Apollo Park visit in Beijing

Contact China Innovation Visits with the preferred date, group size and learning priorities to coordinate a Beijing programme suited to the delegation.