Driving SMARTS 2.0 Benchmark Task

AKA: Driving SMARTS Competition 2.0.
Context:
- It can (typically) require the agents capable of driving from a starting point to a destination, rapidly and safely, amid background traffic.
- It can (typically) leverage a SMARTS Simulation Environment and large-scale Naturalistic Driving Data.
- It can be focused on Driving Scenarios, such as: cruising, overtaking, merging, left turns at unsignalized intersections, and situations where another vehicle cuts off the agent.
- It can reference metrics such as safety, smoothness of driving, task completion (percentage of completed scenarios), violation of traffic rules, and the time taken to complete the task.
- …
Example(s):
- …
Counter-Example(s):
- Manual driving tasks and evaluations.
See: Autonomous Driving, SMARTS Simulation Environment, Naturalistic Driving Data.

References

(Driving SMARTS Competition 2.0, 2023) ⇒ https://smarts-project.github.io/competition/2023_driving_smarts/ Retrieved:2023-7-9.
- Autonomous driving (AD) is the next frontier of artificial intelligence and machine learning. Intending to further research in AD, we invite you to participate in an autonomous driving competition.
- This competition seeks to advance autonomous driving by developing agents that can drive as quickly and safely as possible from the start to destination amid background traffic. Data for the competition consists of large-scale naturalistic driving data replayed within SMARTS simulation environment. The following typical driving scenarios are tested: cruising, overtaking, merging, left turns at unsignalized intersections and being cut off by another vehicle. These scenarios are mined from the naturalistic data, manipulated and replayed in SMARTS. For some scenarios, interactive background vehicles are added in SMARTS.
- Agents will be ranked according to metrics on safety and comfort (smoothness and safe driving), task completion (% of completed scenarios), traffic rule violation, and completion time.