Vectorized Environment
An environment-wrapping pattern in reinforcement learning that batches multiple independent environment instances behind a unified interface, advancing them in parallel and returning batched observations and rewards to increase sampling throughput.
Timeline (last 90 days)
A vectorized environment wraps multiple standard environment instances into a batched interface, presents itself externally as one environment, internally advances N independent copies in parallel, and returns a batch of observations, rewards, and done flags each time.
Vectorized environments let sampling throughput grow linearly with the degree of parallelism while maintaining the physical isolation of each copy, providing policy gradient methods with low-variance, independent and identically distributed batch experience.
Using vectorized environments for parallel sampling can maximize data throughput on a single machine.
All Facts (3)
A vectorized environment wraps multiple standard environment instances into a batched interface, presents itself externally as one environment, internally advances N independent copies in parallel, and returns a batch of observations, rewards, and done flags each time.
50%UnverifiedVectorized environments let sampling throughput grow linearly with the degree of parallelism while maintaining the physical isolation of each copy, providing policy gradient methods with low-variance, independent and identically distributed batch experience.
50%UnverifiedUsing vectorized environments for parallel sampling can maximize data throughput on a single machine.
50%