> For the complete documentation index, see [llms.txt](https://slm-lab.gitbook.io/slm-lab/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://slm-lab.gitbook.io/slm-lab/benchmark-results/discrete-benchmark.md).

# Discrete Benchmark

## Classic Control & Box2D Results

SLM Lab v5.2 validates algorithms on [Gymnasium](https://gymnasium.farama.org/) discrete environments using the TorchArc architecture. These benchmarks cover:

* [**Classic Control**](https://gymnasium.farama.org/environments/classic_control/): CartPole, Acrobot, Pendulum—simple physics tasks ideal for algorithm validation
* [**Box2D**](https://gymnasium.farama.org/environments/box2d/): LunarLander—2D physics with more complex dynamics

Results below are from February–March 2026 benchmark runs using Gymnasium v5 environments. TorchArc specs for existing algorithms; CrossQ uses standard MLP specs.

All trained models and metrics are publicly available on [HuggingFace](https://huggingface.co/datasets/SLM-Lab/benchmark).

### Methodology

Results show **Trial-level** performance:

1. **Trial** = 4 Sessions with different random seeds
2. **Session** = One complete training run
3. **Score** = Final 100-checkpoint moving average (`total_reward_ma`)

The trial score is the mean across 4 sessions, providing statistically meaningful results.

### Standardized Settings

| Category        | num\_envs | max\_frame | log\_frequency | ASHA grace\_period |
| --------------- | --------- | ---------- | -------------- | ------------------ |
| Classic Control | 4         | 2e5-3e5    | 500            | 1e4                |
| Box2D           | 8         | 3e5        | 1000           | 5e4                |

The `grace_period` is the minimum frames before ASHA early stopping can terminate underperforming trials.

{% hint style="warning" %}
**v5 vs v4 Difficulty:** Gymnasium environments have stricter termination and reward handling:

* **LunarLander-v3** is notably harder than v2—stricter landing criteria, lower typical scores
* **Pendulum-v1** uses different reward scaling than v0
* Expect **5-15% lower scores** compared to OpenAI Gym benchmarks

See [Gymnasium docs](https://gymnasium.farama.org/) for environment-specific changes.
{% endhint %}

### Running Benchmarks

**Local** - runs on your machine (Classic Control completes in minutes on CPU):

```bash
slm-lab run slm_lab/spec/benchmark_arc/ppo/ppo_classic_arc.yaml ppo_cartpole_arc train
slm-lab run slm_lab/spec/benchmark_arc/dqn/dqn_box2d_arc.yaml ddqn_per_concat_lunar_arc train
```

**Remote** - cloud GPU via [dstack](https://dstack.ai), auto-syncs to HuggingFace:

```bash
source .env && slm-lab run-remote slm_lab/spec/benchmark_arc/ppo/ppo_classic_arc.yaml ppo_cartpole_arc train -n cartpole
source .env && slm-lab run-remote slm_lab/spec/benchmark_arc/dqn/dqn_box2d_arc.yaml ddqn_per_concat_lunar_arc train -n lunar
```

Remote setup: `cp .env.example .env` then set `HF_TOKEN`. See [Remote Training](/slm-lab/using-slm-lab/remote-training.md) for dstack config.

{% hint style="info" %}
**GPU not required for Classic Control.** These environments train fast on CPU. Box2D (LunarLander) benefits from GPU but still runs fine locally.
{% endhint %}

### Download and Replay

```bash
# List all available experiments (requires HF_REPO=SLM-Lab/benchmark in .env)
source .env && slm-lab list

# Download a specific experiment
source .env && slm-lab pull ppo_cartpole_arc

# Replay the trained agent
slm-lab run slm_lab/spec/benchmark_arc/ppo/ppo_classic_arc.yaml ppo_cartpole_arc enjoy@data/ppo_cartpole_arc_*/ppo_cartpole_arc_t0_spec.yaml
```

***

## Results

### Classic Control

#### CartPole-v1

[Docs](https://gymnasium.farama.org/environments/classic_control/cart_pole/) | State: Box(4) | Action: Discrete(2) | Target: >400

**Settings**: max\_frame 2e5 | num\_envs 4 | max\_session 4 | log\_frequency 500

| Algorithm | Status | MA     | SPEC\_FILE                                                                                                                                                         | SPEC\_NAME                          | HF Data                                                                                                                                                                         |
| --------- | ------ | ------ | ------------------------------------------------------------------------------------------------------------------------------------------------------------------ | ----------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| REINFORCE | ✅      | 483.31 | [slm\_lab/spec/benchmark\_arc/reinforce/reinforce\_arc.yaml](https://github.com/kengz/SLM-Lab/blob/master/slm_lab/spec/benchmark_arc/reinforce/reinforce_arc.yaml) | reinforce\_cartpole\_arc            | [reinforce\_cartpole\_arc\_2026\_02\_11\_135616](https://huggingface.co/datasets/SLM-Lab/benchmark/tree/main/data/reinforce_cartpole_arc_2026_02_11_135616)                     |
| SARSA     | ✅      | 430.95 | [slm\_lab/spec/benchmark\_arc/sarsa/sarsa\_arc.yaml](https://github.com/kengz/SLM-Lab/blob/master/slm_lab/spec/benchmark_arc/sarsa/sarsa_arc.yaml)                 | sarsa\_boltzmann\_cartpole\_arc     | [sarsa\_boltzmann\_cartpole\_arc\_2026\_02\_11\_135616](https://huggingface.co/datasets/SLM-Lab/benchmark/tree/main/data/sarsa_boltzmann_cartpole_arc_2026_02_11_135616)        |
| DQN       | ⚠️     | 239.94 | [slm\_lab/spec/benchmark\_arc/dqn/dqn\_classic\_arc.yaml](https://github.com/kengz/SLM-Lab/blob/master/slm_lab/spec/benchmark_arc/dqn/dqn_classic_arc.yaml)        | dqn\_boltzmann\_cartpole\_arc       | [dqn\_boltzmann\_cartpole\_arc\_2026\_02\_11\_135648](https://huggingface.co/datasets/SLM-Lab/benchmark/tree/main/data/dqn_boltzmann_cartpole_arc_2026_02_11_135648)            |
| DDQN+PER  | ✅      | 451.51 | [slm\_lab/spec/benchmark\_arc/dqn/dqn\_classic\_arc.yaml](https://github.com/kengz/SLM-Lab/blob/master/slm_lab/spec/benchmark_arc/dqn/dqn_classic_arc.yaml)        | ddqn\_per\_boltzmann\_cartpole\_arc | [ddqn\_per\_boltzmann\_cartpole\_arc\_2026\_02\_11\_140518](https://huggingface.co/datasets/SLM-Lab/benchmark/tree/main/data/ddqn_per_boltzmann_cartpole_arc_2026_02_11_140518) |
| A2C       | ✅      | 496.68 | [slm\_lab/spec/benchmark\_arc/a2c/a2c\_classic\_arc.yaml](https://github.com/kengz/SLM-Lab/blob/master/slm_lab/spec/benchmark_arc/a2c/a2c_classic_arc.yaml)        | a2c\_gae\_cartpole\_arc             | [a2c\_gae\_cartpole\_arc\_2026\_02\_11\_142531](https://huggingface.co/datasets/SLM-Lab/benchmark/tree/main/data/a2c_gae_cartpole_arc_2026_02_11_142531)                        |
| PPO       | ✅      | 498.94 | [slm\_lab/spec/benchmark\_arc/ppo/ppo\_classic\_arc.yaml](https://github.com/kengz/SLM-Lab/blob/master/slm_lab/spec/benchmark_arc/ppo/ppo_classic_arc.yaml)        | ppo\_cartpole\_arc                  | [ppo\_cartpole\_arc\_2026\_02\_11\_144029](https://huggingface.co/datasets/SLM-Lab/benchmark/tree/main/data/ppo_cartpole_arc_2026_02_11_144029)                                 |
| SAC       | ✅      | 406.09 | [slm\_lab/spec/benchmark\_arc/sac/sac\_classic\_arc.yaml](https://github.com/kengz/SLM-Lab/blob/master/slm_lab/spec/benchmark_arc/sac/sac_classic_arc.yaml)        | sac\_cartpole\_arc                  | [sac\_cartpole\_arc\_2026\_02\_11\_144155](https://huggingface.co/datasets/SLM-Lab/benchmark/tree/main/data/sac_cartpole_arc_2026_02_11_144155)                                 |
| CrossQ    | ⚠️     | 334.59 | [slm\_lab/spec/benchmark/crossq/crossq\_classic.yaml](https://github.com/kengz/SLM-Lab/blob/master/slm_lab/spec/benchmark/crossq/crossq_classic.yaml)              | crossq\_cartpole                    | [crossq\_cartpole\_2026\_03\_02\_100434](https://huggingface.co/datasets/SLM-Lab/benchmark/tree/main/data/crossq_cartpole_2026_03_02_100434)                                    |

![CartPole-v1](https://huggingface.co/datasets/SLM-Lab/benchmark/resolve/v5.2.0/docs/plots/CartPole-v1_multi_trial_graph_mean_returns_ma_vs_frames.png)

#### Acrobot-v1

[Docs](https://gymnasium.farama.org/environments/classic_control/acrobot/) | State: Box(6) | Action: Discrete(3) | Target: >-100

**Settings**: max\_frame 3e5 | num\_envs 4 | max\_session 4 | log\_frequency 500

| Algorithm | Status | MA      | SPEC\_FILE                                                                                                                                                  | SPEC\_NAME                   | HF Data                                                                                                                                                            |
| --------- | ------ | ------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------- | ---------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------ |
| DQN       | ✅      | -94.17  | [slm\_lab/spec/benchmark\_arc/dqn/dqn\_classic\_arc.yaml](https://github.com/kengz/SLM-Lab/blob/master/slm_lab/spec/benchmark_arc/dqn/dqn_classic_arc.yaml) | dqn\_boltzmann\_acrobot\_arc | [dqn\_boltzmann\_acrobot\_arc\_2026\_02\_11\_144342](https://huggingface.co/datasets/SLM-Lab/benchmark/tree/main/data/dqn_boltzmann_acrobot_arc_2026_02_11_144342) |
| DDQN+PER  | ✅      | -83.92  | [slm\_lab/spec/benchmark\_arc/dqn/dqn\_classic\_arc.yaml](https://github.com/kengz/SLM-Lab/blob/master/slm_lab/spec/benchmark_arc/dqn/dqn_classic_arc.yaml) | ddqn\_per\_acrobot\_arc      | [ddqn\_per\_acrobot\_arc\_2026\_02\_11\_153725](https://huggingface.co/datasets/SLM-Lab/benchmark/tree/main/data/ddqn_per_acrobot_arc_2026_02_11_153725)           |
| A2C       | ✅      | -83.99  | [slm\_lab/spec/benchmark\_arc/a2c/a2c\_classic\_arc.yaml](https://github.com/kengz/SLM-Lab/blob/master/slm_lab/spec/benchmark_arc/a2c/a2c_classic_arc.yaml) | a2c\_gae\_acrobot\_arc       | [a2c\_gae\_acrobot\_arc\_2026\_02\_11\_153806](https://huggingface.co/datasets/SLM-Lab/benchmark/tree/main/data/a2c_gae_acrobot_arc_2026_02_11_153806)             |
| PPO       | ✅      | -81.28  | [slm\_lab/spec/benchmark\_arc/ppo/ppo\_classic\_arc.yaml](https://github.com/kengz/SLM-Lab/blob/master/slm_lab/spec/benchmark_arc/ppo/ppo_classic_arc.yaml) | ppo\_acrobot\_arc            | [ppo\_acrobot\_arc\_2026\_02\_11\_153758](https://huggingface.co/datasets/SLM-Lab/benchmark/tree/main/data/ppo_acrobot_arc_2026_02_11_153758)                      |
| SAC       | ✅      | -92.60  | [slm\_lab/spec/benchmark\_arc/sac/sac\_classic\_arc.yaml](https://github.com/kengz/SLM-Lab/blob/master/slm_lab/spec/benchmark_arc/sac/sac_classic_arc.yaml) | sac\_acrobot\_arc            | [sac\_acrobot\_arc\_2026\_02\_11\_162211](https://huggingface.co/datasets/SLM-Lab/benchmark/tree/main/data/sac_acrobot_arc_2026_02_11_162211)                      |
| CrossQ    | ✅      | -103.13 | [slm\_lab/spec/benchmark/crossq/crossq\_classic.yaml](https://github.com/kengz/SLM-Lab/blob/master/slm_lab/spec/benchmark/crossq/crossq_classic.yaml)       | crossq\_acrobot              | [crossq\_acrobot\_2026\_02\_23\_153622](https://huggingface.co/datasets/SLM-Lab/benchmark/tree/main/data/crossq_acrobot_2026_02_23_153622)                         |

![Acrobot-v1](https://huggingface.co/datasets/SLM-Lab/benchmark/resolve/v5.2.0/docs/plots/Acrobot-v1_multi_trial_graph_mean_returns_ma_vs_frames.png)

#### Pendulum-v1

[Docs](https://gymnasium.farama.org/environments/classic_control/pendulum/) | State: Box(3) | Action: Box(1) | Target: >-200

**Settings**: max\_frame 3e5 | num\_envs 4 | max\_session 4 | log\_frequency 500

| Algorithm | Status | MA      | SPEC\_FILE                                                                                                                                                  | SPEC\_NAME              | HF Data                                                                                                                                                  |
| --------- | ------ | ------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------- | ----------------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------- |
| A2C       | ❌      | -820.74 | [slm\_lab/spec/benchmark\_arc/a2c/a2c\_classic\_arc.yaml](https://github.com/kengz/SLM-Lab/blob/master/slm_lab/spec/benchmark_arc/a2c/a2c_classic_arc.yaml) | a2c\_gae\_pendulum\_arc | [a2c\_gae\_pendulum\_arc\_2026\_02\_11\_162217](https://huggingface.co/datasets/SLM-Lab/benchmark/tree/main/data/a2c_gae_pendulum_arc_2026_02_11_162217) |
| PPO       | ✅      | -174.87 | [slm\_lab/spec/benchmark\_arc/ppo/ppo\_classic\_arc.yaml](https://github.com/kengz/SLM-Lab/blob/master/slm_lab/spec/benchmark_arc/ppo/ppo_classic_arc.yaml) | ppo\_pendulum\_arc      | [ppo\_pendulum\_arc\_2026\_02\_11\_162156](https://huggingface.co/datasets/SLM-Lab/benchmark/tree/main/data/ppo_pendulum_arc_2026_02_11_162156)          |
| SAC       | ✅      | -150.97 | [slm\_lab/spec/benchmark\_arc/sac/sac\_classic\_arc.yaml](https://github.com/kengz/SLM-Lab/blob/master/slm_lab/spec/benchmark_arc/sac/sac_classic_arc.yaml) | sac\_pendulum\_arc      | [sac\_pendulum\_arc\_2026\_02\_11\_162240](https://huggingface.co/datasets/SLM-Lab/benchmark/tree/main/data/sac_pendulum_arc_2026_02_11_162240)          |
| CrossQ    | ✅      | -145.66 | [slm\_lab/spec/benchmark/crossq/crossq\_classic.yaml](https://github.com/kengz/SLM-Lab/blob/master/slm_lab/spec/benchmark/crossq/crossq_classic.yaml)       | crossq\_pendulum        | [crossq\_pendulum\_2026\_02\_28\_130648](https://huggingface.co/datasets/SLM-Lab/benchmark/tree/main/data/crossq_pendulum_2026_02_28_130648)             |

![Pendulum-v1](https://huggingface.co/datasets/SLM-Lab/benchmark/resolve/v5.2.0/docs/plots/Pendulum-v1_multi_trial_graph_mean_returns_ma_vs_frames.png)

### Box2D

#### LunarLander-v3 (Discrete)

[Docs](https://gymnasium.farama.org/environments/box2d/lunar_lander/) | State: Box(8) | Action: Discrete(4) | Target: >200

**Settings**: max\_frame 3e5 | num\_envs 8 | max\_session 4 | log\_frequency 1000

| Algorithm | Status | MA     | SPEC\_FILE                                                                                                                                                  | SPEC\_NAME                    | HF Data                                                                                                                                                             |
| --------- | ------ | ------ | ----------------------------------------------------------------------------------------------------------------------------------------------------------- | ----------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| DQN       | ⚠️     | 195.21 | [slm\_lab/spec/benchmark\_arc/dqn/dqn\_box2d\_arc.yaml](https://github.com/kengz/SLM-Lab/blob/master/slm_lab/spec/benchmark_arc/dqn/dqn_box2d_arc.yaml)     | dqn\_concat\_lunar\_arc       | [dqn\_concat\_lunar\_arc\_2026\_02\_11\_201407](https://huggingface.co/datasets/SLM-Lab/benchmark/tree/main/data/dqn_concat_lunar_arc_2026_02_11_201407)            |
| DDQN+PER  | ✅      | 265.90 | [slm\_lab/spec/benchmark\_arc/dqn/dqn\_box2d\_arc.yaml](https://github.com/kengz/SLM-Lab/blob/master/slm_lab/spec/benchmark_arc/dqn/dqn_box2d_arc.yaml)     | ddqn\_per\_concat\_lunar\_arc | [ddqn\_per\_concat\_lunar\_arc\_2026\_02\_13\_105115](https://huggingface.co/datasets/SLM-Lab/benchmark/tree/main/data/ddqn_per_concat_lunar_arc_2026_02_13_105115) |
| A2C       | ❌      | 27.38  | [slm\_lab/spec/benchmark\_arc/a2c/a2c\_classic\_arc.yaml](https://github.com/kengz/SLM-Lab/blob/master/slm_lab/spec/benchmark_arc/a2c/a2c_classic_arc.yaml) | a2c\_gae\_lunar\_arc          | [a2c\_gae\_lunar\_arc\_2026\_02\_11\_224304](https://huggingface.co/datasets/SLM-Lab/benchmark/tree/main/data/a2c_gae_lunar_arc_2026_02_11_224304)                  |
| PPO       | ⚠️     | 183.30 | [slm\_lab/spec/benchmark\_arc/ppo/ppo\_box2d\_arc.yaml](https://github.com/kengz/SLM-Lab/blob/master/slm_lab/spec/benchmark_arc/ppo/ppo_box2d_arc.yaml)     | ppo\_lunar\_arc               | [ppo\_lunar\_arc\_2026\_02\_11\_201303](https://huggingface.co/datasets/SLM-Lab/benchmark/tree/main/data/ppo_lunar_arc_2026_02_11_201303)                           |
| SAC       | ⚠️     | 106.17 | [slm\_lab/spec/benchmark\_arc/sac/sac\_box2d\_arc.yaml](https://github.com/kengz/SLM-Lab/blob/master/slm_lab/spec/benchmark_arc/sac/sac_box2d_arc.yaml)     | sac\_lunar\_arc               | [sac\_lunar\_arc\_2026\_02\_11\_201417](https://huggingface.co/datasets/SLM-Lab/benchmark/tree/main/data/sac_lunar_arc_2026_02_11_201417)                           |
| CrossQ    | ❌      | 139.21 | [slm\_lab/spec/benchmark/crossq/crossq\_box2d.yaml](https://github.com/kengz/SLM-Lab/blob/master/slm_lab/spec/benchmark/crossq/crossq_box2d.yaml)           | crossq\_lunar                 | [crossq\_lunar\_2026\_02\_28\_130733](https://huggingface.co/datasets/SLM-Lab/benchmark/tree/main/data/crossq_lunar_2026_02_28_130733)                              |

![LunarLander-v3 Discrete](https://huggingface.co/datasets/SLM-Lab/benchmark/resolve/v5.2.0/docs/plots/LunarLander-v3_Discrete_multi_trial_graph_mean_returns_ma_vs_frames.png)

#### LunarLander-v3 (Continuous)

[Docs](https://gymnasium.farama.org/environments/box2d/lunar_lander/) | State: Box(8) | Action: Box(2) | Target: >200

**Settings**: max\_frame 3e5 | num\_envs 8 | max\_session 4 | log\_frequency 1000

| Algorithm | Status | MA     | SPEC\_FILE                                                                                                                                                  | SPEC\_NAME                       | HF Data                                                                                                                                                                   |
| --------- | ------ | ------ | ----------------------------------------------------------------------------------------------------------------------------------------------------------- | -------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| A2C       | ❌      | -76.81 | [slm\_lab/spec/benchmark\_arc/a2c/a2c\_classic\_arc.yaml](https://github.com/kengz/SLM-Lab/blob/master/slm_lab/spec/benchmark_arc/a2c/a2c_classic_arc.yaml) | a2c\_gae\_lunar\_continuous\_arc | [a2c\_gae\_lunar\_continuous\_arc\_2026\_02\_11\_224301](https://huggingface.co/datasets/SLM-Lab/benchmark/tree/main/data/a2c_gae_lunar_continuous_arc_2026_02_11_224301) |
| PPO       | ⚠️     | 132.58 | [slm\_lab/spec/benchmark\_arc/ppo/ppo\_box2d\_arc.yaml](https://github.com/kengz/SLM-Lab/blob/master/slm_lab/spec/benchmark_arc/ppo/ppo_box2d_arc.yaml)     | ppo\_lunar\_continuous\_arc      | [ppo\_lunar\_continuous\_arc\_2026\_02\_11\_224229](https://huggingface.co/datasets/SLM-Lab/benchmark/tree/main/data/ppo_lunar_continuous_arc_2026_02_11_224229)          |
| SAC       | ⚠️     | 125.00 | [slm\_lab/spec/benchmark\_arc/sac/sac\_box2d\_arc.yaml](https://github.com/kengz/SLM-Lab/blob/master/slm_lab/spec/benchmark_arc/sac/sac_box2d_arc.yaml)     | sac\_lunar\_continuous\_arc      | [sac\_lunar\_continuous\_arc\_2026\_02\_12\_222203](https://huggingface.co/datasets/SLM-Lab/benchmark/tree/main/data/sac_lunar_continuous_arc_2026_02_12_222203)          |
| CrossQ    | ✅      | 268.91 | [slm\_lab/spec/benchmark/crossq/crossq\_box2d.yaml](https://github.com/kengz/SLM-Lab/blob/master/slm_lab/spec/benchmark/crossq/crossq_box2d.yaml)           | crossq\_lunar\_continuous        | [crossq\_lunar\_continuous\_2026\_03\_01\_140517](https://huggingface.co/datasets/SLM-Lab/benchmark/tree/main/data/crossq_lunar_continuous_2026_03_01_140517)             |

![LunarLander-v3 Continuous](https://huggingface.co/datasets/SLM-Lab/benchmark/resolve/v5.2.0/docs/plots/LunarLander-v3_Continuous_multi_trial_graph_mean_returns_ma_vs_frames.png)

**Legend:** ✅ Solved | ⚠️ Close (>80%) | ❌ Failed

***

## Historical Results (v4)

<details>

<summary>OpenAI Gym Results (v4) - click to expand</summary>

{% hint style="info" %}
These results from SLM Lab v4 used OpenAI Gym environments (now deprecated). Environment versions differ from current Gymnasium versions. Unity environments are no longer included in the core package.
{% endhint %}

* [Upload PR #427](https://github.com/kengz/SLM-Lab/pull/427)
* [Google Drive data](https://drive.google.com/file/d/1lb9Hn22Uzb67ndotRULiUspk3TnmKp6Z/view?usp=sharing)

| Env. \ Alg. | DQN   | DDQN+PER | A2C (GAE) | A2C (n-step) | PPO       | SAC     |
| ----------- | ----- | -------- | --------- | ------------ | --------- | ------- |
| Breakout    | 80.88 | 182      | 377       | 398          | **443**   | 3.51\*  |
| Pong        | 18.48 | 20.5     | 19.31     | 19.56        | **20.58** | 19.87\* |
| Qbert       | 5494  | 11426    | 12405     | **13590**    | 13460     | 923\*   |
| Seaquest    | 1185  | **4405** | 1070      | 1684         | 1715      | 171\*   |
| LunarLander | 192   | 233      | 25.21     | 68.23        | 214       | **276** |

> Episode score at the end of training. Reported scores are the average over the last 100 checkpoints, averaged over 4 Sessions. Results marked with `*` used async SAC.

</details>

For the full Atari benchmark, see [Atari Benchmark](/slm-lab/benchmark-results/atari-benchmark.md).