Randomized Ensembled Double Q-Learning: Learning Fast Without a Model
About
Using a high Update-To-Data (UTD) ratio, model-based methods have recently achieved much higher sample efficiency than previous model-free methods for continuous-action DRL benchmarks. In this paper, we introduce a simple model-free algorithm, Randomized Ensembled Double Q-Learning (REDQ), and show that its performance is just as good as, if not better than, a state-of-the-art model-based algorithm for the MuJoCo benchmark. Moreover, REDQ can achieve this performance using fewer parameters than the model-based method, and with less wall-clock run time. REDQ has three carefully integrated ingredients which allow it to achieve its high performance: (i) a UTD ratio >> 1; (ii) an ensemble of Q functions; (iii) in-target minimization across a random subset of Q functions from the ensemble. Through carefully designed experiments, we provide a detailed analysis of REDQ and related model-free algorithms. To our knowledge, REDQ is the first successful model-free DRL algorithm for continuous-action spaces using a UTD ratio >> 1.
Related benchmarks
| Task | Dataset | Result | Rank | |
|---|---|---|---|---|
| Continuous Control | MuJoCo Ant v4 | Average Return5.31e+3 | 46 | |
| Continuous Control | MuJoCo Walker2d v4 | -- | 39 | |
| Continuous Control | MuJoCo HalfCheetah v4 | Average Return1.15e+4 | 36 | |
| Continuous Control | MuJoCo v5 | Ant Score4.83e+3 | 15 | |
| Continuous Control | DeepMind Control Suite (DMC) | Cheetah Run866 | 15 | |
| Continuous Control | Gym MuJoCo Hopper v4 | Average Return3.30e+3 | 15 | |
| Continuous Control | Gym MuJoCo Suite Aggregate | IQM1.135 | 15 | |
| Continuous Control | Gym MuJoCo Humanoid v4 | Average Return5.28e+3 | 15 | |
| Continuous Control | Mujoco | Ant-v54.83e+3 | 9 | |
| Continuous Control | OpenAI Gym Mujoco 100K steps v2 (train) | InvertedPendulum-v2 Score1.00e+3 | 5 |