Alibaba launches the world's first autonomous driving simulation road testing platform capable of testing 8 million kilometers a day

On April 22, reporters learned that Alibaba DAMO Academy released the world's first "hybrid simulation test platform" for autonomous driving. The platform uses simulation technology that combines virtual and real life, introduces real road test scenarios and cloud trainers, and only takes 30 seconds to simulate an extreme scenario. The system's daily virtual test mileage can exceed 8 million kilometers, greatly improving the efficiency of autonomous driving AI model training. This technology will accelerate autonomous driving to the L5 stage.

Road testing is the core link of autonomous driving. Research shows that autonomous driving cars need to accumulate 17.7 billion kilometers of test data to ensure the safety of the entire link of autonomous driving perception, decision-making, and control. Traditional pure virtual simulation test platforms can quickly complete autonomous driving road test mileage, but they still face the key problem of low efficiency in extreme scene training: insufficient extreme scene data cannot restore the uncertainty of real road conditions, the system cannot accurately respond to emergencies in real road conditions, and autonomous driving is difficult to achieve further breakthroughs.

DAMO Academy's pioneering hybrid simulation test platform for autonomous driving solves this problem. The platform bridges the gap between the online virtual fixed environment and the uncertainty of offline real road conditions. Traditional simulation platforms find it difficult to simulate random human intervention through algorithms, but on DAMO Academy's platform, not only can real road test data be used to automatically generate simulation scenarios, but also random human intervention can be used to simulate acceleration, sharp turns, emergency stops and other scenarios in real time, making obstacle avoidance training for autonomous vehicles more difficult.

To address the problem of insufficient data for extreme scenarios, the platform can arbitrarily increase extreme road test scenario variables. In actual road tests, it may take a month to reproduce an extreme scenario, but the platform can complete the construction and testing of special scenarios such as rainy and snowy weather and poor lighting conditions at night within 30 seconds, and the number of scenario constructions that can be supported daily is in the millions.

Industry experts pointed out that this platform solves the problem of reproducing extreme scenarios on a large scale, increasing the training efficiency of these key scenarios by millions of times, which will accelerate autonomous driving towards the L5 stage.