gmgn 0.4.3

A reinforcement learning environments library for Rust.
Documentation
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
//! Two-link acrobot swing-up environment.
//!
//! The system consists of two links connected linearly to form a chain, with
//! one end fixed. The joint between the two links is actuated. The goal is to
//! swing the free end above a target height.
//!
//! Based on Sutton's "Generalization in Reinforcement Learning" (`NeurIPS` 1995).
//!
//! Mirrors [Gymnasium `Acrobot-v1`](https://gymnasium.farama.org/environments/classic_control/acrobot/).

use std::collections::HashMap;
use std::f64::consts::PI;

use rand::RngExt as _;

use crate::env::{Env, RenderFrame, RenderMode, ResetResult, StepResult};
use crate::error::{Error, Result};
#[cfg(feature = "render")]
use crate::render::{Canvas, RenderWindow};
use crate::rng::{self, Rng};
use crate::space::{BoundedSpace, Discrete, Space};

const LINK_LENGTH_1: f64 = 1.0;
const LINK_LENGTH_2: f64 = 1.0;
const LINK_MASS_1: f64 = 1.0;
const LINK_MASS_2: f64 = 1.0;
const LINK_COM_POS_1: f64 = 0.5;
const LINK_COM_POS_2: f64 = 0.5;
const LINK_MOI: f64 = 1.0;
const GRAVITY: f64 = 9.8;
const DT: f64 = 0.2;
const MAX_VEL_1: f64 = 4.0 * PI;
const MAX_VEL_2: f64 = 9.0 * PI;

#[cfg(feature = "render")]
const SCREEN_DIM: u32 = 500;
#[cfg(feature = "render")]
const RENDER_FPS: usize = 15;

/// Available torques mapped from discrete actions {0, 1, 2}.
const AVAIL_TORQUE: [f64; 3] = [-1.0, 0.0, 1.0];

/// Configuration for [`AcrobotEnv`].
#[derive(Debug, Clone, Copy)]
pub struct AcrobotConfig {
    /// The render mode for this environment.
    pub render_mode: RenderMode,
}

impl Default for AcrobotConfig {
    fn default() -> Self {
        Self {
            render_mode: RenderMode::None,
        }
    }
}

/// Wrap angle to `[-π, π]`.
fn wrap(x: f64, lo: f64, hi: f64) -> f64 {
    let range = hi - lo;
    ((x - lo) % range + range) % range + lo
}

/// Clamp value to `[lo, hi]`.
const fn bound(x: f64, lo: f64, hi: f64) -> f64 {
    x.clamp(lo, hi)
}

/// Compute the state derivatives for the acrobot (book dynamics).
///
/// `s_aug = [theta1, theta2, dtheta1, dtheta2, torque]`
fn dsdt(s_aug: &[f64; 5]) -> [f64; 5] {
    let [theta1, theta2, dtheta1, dtheta2, a] = *s_aug;

    let m1 = LINK_MASS_1;
    let m2 = LINK_MASS_2;
    let l1 = LINK_LENGTH_1;
    let lc1 = LINK_COM_POS_1;
    let lc2 = LINK_COM_POS_2;
    let i1 = LINK_MOI;
    let i2 = LINK_MOI;
    let g = GRAVITY;

    let d1 = (m1 * lc1).mul_add(
        lc1,
        m2 * (2.0 * l1 * lc2).mul_add(theta2.cos(), l1.mul_add(l1, lc2 * lc2)),
    ) + i1
        + i2;
    let d2 = m2.mul_add(lc2.mul_add(lc2, l1 * lc2 * theta2.cos()), i2);
    let phi2 = m2 * lc2 * g * (theta1 + theta2 - PI / 2.0).cos();
    let phi1 = (m1.mul_add(lc1, m2 * l1) * g).mul_add(
        (theta1 - PI / 2.0).cos(),
        (-m2 * l1 * lc2 * dtheta2 * dtheta2).mul_add(
            theta2.sin(),
            -(2.0 * m2 * l1 * lc2 * dtheta2 * dtheta1 * theta2.sin()),
        ),
    ) + phi2;

    // Book dynamics (Sutton & Barto).
    let ddtheta2 = ((m2 * l1 * lc2 * dtheta1 * dtheta1)
        .mul_add(-theta2.sin(), (d2 / d1).mul_add(phi1, a))
        - phi2)
        / ((m2 * lc2).mul_add(lc2, i2) - d2 * d2 / d1);
    let ddtheta1 = -d2.mul_add(ddtheta2, phi1) / d1;

    [dtheta1, dtheta2, ddtheta1, ddtheta2, 0.0]
}

/// Fourth-order Runge-Kutta integrator for the acrobot dynamics.
fn rk4(state_aug: &[f64; 5], dt: f64) -> [f64; 5] {
    // Single step RK4 over interval [0, dt].
    let k1 = dsdt(state_aug);

    let mut s2 = [0.0; 5];
    for i in 0..5 {
        s2[i] = (0.5 * dt).mul_add(k1[i], state_aug[i]);
    }
    let k2 = dsdt(&s2);

    let mut s3 = [0.0; 5];
    for i in 0..5 {
        s3[i] = (0.5 * dt).mul_add(k2[i], state_aug[i]);
    }
    let k3 = dsdt(&s3);

    let mut s4 = [0.0; 5];
    for i in 0..5 {
        s4[i] = dt.mul_add(k3[i], state_aug[i]);
    }
    let k4 = dsdt(&s4);

    let mut result = [0.0; 5];
    for i in 0..5 {
        result[i] = (dt / 6.0).mul_add(
            2.0f64.mul_add(k3[i], 2.0f64.mul_add(k2[i], k1[i])) + k4[i],
            state_aug[i],
        );
    }
    result
}

/// The Acrobot swing-up environment with discrete actions.
///
/// # Action Space
///
/// [`Discrete(3)`](Discrete): apply torque −1 (0), 0 (1), or +1 (2).
///
/// # Observation Space
///
/// [`BoundedSpace`] of shape `[6]`:
///
/// | Index | Observation          | Min    | Max    |
/// |-------|----------------------|--------|--------|
/// | 0     | cos(θ₁)              | −1     | 1      |
/// | 1     | sin(θ₁)              | −1     | 1      |
/// | 2     | cos(θ₂)              | −1     | 1      |
/// | 3     | sin(θ₂)              | −1     | 1      |
/// | 4     | Angular vel θ₁       | −4π    | 4π     |
/// | 5     | Angular vel θ₂       | −9π    | 9π     |
///
/// # Rewards
///
/// −1 per step until termination; 0 on the terminal step.
///
/// # Episode End
///
/// - **Termination**: `−cos(θ₁) − cos(θ₂ + θ₁) > 1.0`.
/// - **Truncation**: handled externally by a [`TimeLimit`](crate::wrappers::TimeLimit)
///   wrapper (typically 500 steps).
pub struct AcrobotEnv {
    action_space: Discrete,
    observation_space: BoundedSpace,

    /// Internal state: `[theta1, theta2, dtheta1, dtheta2]`.
    state: Option<[f64; 4]>,
    rng: Rng,
    render_mode: RenderMode,

    #[cfg(feature = "render")]
    canvas: Option<Canvas>,
    #[cfg(feature = "render")]
    window: Option<RenderWindow>,
}

impl std::fmt::Debug for AcrobotEnv {
    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
        f.debug_struct("AcrobotEnv")
            .field("state", &self.state)
            .field("render_mode", &self.render_mode)
            .finish_non_exhaustive()
    }
}

impl AcrobotEnv {
    /// Create a new acrobot environment.
    ///
    /// # Errors
    ///
    /// Returns an error if the observation space cannot be constructed.
    #[allow(clippy::cast_possible_truncation)]
    pub fn new(config: AcrobotConfig) -> Result<Self> {
        let obs_high = vec![1.0_f32, 1.0, 1.0, 1.0, MAX_VEL_1 as f32, MAX_VEL_2 as f32];
        let obs_low: Vec<f32> = obs_high.iter().map(|&h| -h).collect();

        Ok(Self {
            action_space: Discrete::new(3),
            observation_space: BoundedSpace::new(obs_low, obs_high)?,
            state: None,
            rng: rng::create_rng(None),
            render_mode: config.render_mode,
            #[cfg(feature = "render")]
            canvas: None,
            #[cfg(feature = "render")]
            window: None,
        })
    }

    /// Build the 6-element observation from internal state.
    #[allow(clippy::cast_possible_truncation)]
    fn observation(&self) -> Vec<f32> {
        let [t1, t2, dt1, dt2] = self.state.expect("state must be initialized");
        vec![
            t1.cos() as f32,
            t1.sin() as f32,
            t2.cos() as f32,
            t2.sin() as f32,
            dt1 as f32,
            dt2 as f32,
        ]
    }

    /// Check the termination condition.
    fn is_terminal(&self) -> bool {
        let [t1, t2, _, _] = self.state.expect("state must be initialized");
        -t1.cos() - (t2 + t1).cos() > 1.0
    }

    /// Render the acrobot scene to the internal canvas.
    ///
    /// Matches Gymnasium's rendering: rotated rectangle links with joint
    /// circles, proper colors (cyan links, yellow joints), and correct
    /// coordinate transforms accounting for Gymnasium's Y-flip.
    #[cfg(feature = "render")]
    #[allow(clippy::cast_possible_truncation)]
    fn render_pixels(&mut self) -> Result<RenderFrame> {
        if self.state.is_none() {
            return Err(Error::ResetNeeded { method: "render" });
        }
        let [t1, t2, _, _] = self.state.expect("checked above");

        let canvas = self
            .canvas
            .get_or_insert_with(|| Canvas::new(SCREEN_DIM, SCREEN_DIM));

        canvas.clear(tiny_skia::Color::WHITE);

        let bound = (LINK_LENGTH_1 + LINK_LENGTH_2 + 0.2) as f32;
        let scale = SCREEN_DIM as f32 / (bound * 2.0);
        let offset = SCREEN_DIM as f32 / 2.0;
        let link_width = 0.1 * scale; // half-width for link rectangles

        let link_color = tiny_skia::Color::from_rgba8(0, 204, 204, 255);
        let joint_color = tiny_skia::Color::from_rgba8(204, 204, 0, 255);
        let joint_radius = 0.1 * scale;

        // Joint positions (after Gymnasium's Y-flip transform):
        // Pivot at (offset, offset).
        // Elbow: (L1*sin(t1)*scale + offset, L1*cos(t1)*scale + offset).
        let p1_x = (LINK_LENGTH_1 as f32 * (t1 as f32).sin()).mul_add(scale, offset);
        let p1_y = (LINK_LENGTH_1 as f32 * (t1 as f32).cos()).mul_add(scale, offset);

        // Tip: ((L1*sin(t1)+L2*sin(t1+t2))*scale + offset, ...).
        let p2_x = (LINK_LENGTH_1 as f32)
            .mul_add(
                (t1 as f32).sin(),
                LINK_LENGTH_2 as f32 * ((t1 + t2) as f32).sin(),
            )
            .mul_add(scale, offset);
        let p2_y = (LINK_LENGTH_1 as f32)
            .mul_add(
                (t1 as f32).cos(),
                LINK_LENGTH_2 as f32 * ((t1 + t2) as f32).cos(),
            )
            .mul_add(scale, offset);

        // Target height line at y = offset - scale (1 link-length above pivot).
        let target_y = offset - scale;
        canvas.stroke_line(
            0.0,
            target_y,
            SCREEN_DIM as f32,
            target_y,
            1.0,
            tiny_skia::Color::BLACK,
        );

        // Helper: draw a rotated rectangle link from joint (jx, jy) with
        // rotation angle theta (Gymnasium uses theta - pi/2 as rotate_rad arg).
        // After the combined rotate + Y-flip transform, for local corner (lx, ly):
        //   sx = lx*sin(theta) + ly*cos(theta) + jx
        //   sy = lx*cos(theta) - ly*sin(theta) + jy
        let draw_link = |canvas: &mut Canvas, jx: f32, jy: f32, theta: f64, llen: f32| {
            let sin_t = (theta as f32).sin();
            let cos_t = (theta as f32).cos();
            let lw = link_width;
            // Rectangle: (0, -lw) to (llen, lw)
            let corners: [(f32, f32); 4] = [(0.0, -lw), (0.0, lw), (llen, lw), (llen, -lw)];
            let transformed: Vec<(f32, f32)> = corners
                .iter()
                .map(|&(lx, ly)| {
                    let sx = lx.mul_add(sin_t, ly * cos_t) + jx;
                    let sy = lx.mul_add(cos_t, -(ly * sin_t)) + jy;
                    (sx, sy)
                })
                .collect();
            canvas.fill_polygon(&transformed, link_color);
        };

        // Link 1: from pivot, rotation angle = t1
        let l1_len = LINK_LENGTH_1 as f32 * scale;
        draw_link(canvas, offset, offset, t1, l1_len);

        // Link 2: from elbow, rotation angle = t1 + t2
        let l2_len = LINK_LENGTH_2 as f32 * scale;
        draw_link(canvas, p1_x, p1_y, t1 + t2, l2_len);

        // Joint circles (yellow).
        canvas.fill_circle(offset, offset, joint_radius, joint_color);
        canvas.fill_circle(p1_x, p1_y, joint_radius, joint_color);

        // Tip indicator (small circle).
        canvas.fill_circle(p2_x, p2_y, joint_radius * 0.5, joint_color);

        match self.render_mode {
            RenderMode::Human => {
                let window = self.window.get_or_insert_with(|| {
                    RenderWindow::new(
                        "Acrobot \u{2014} gmgn",
                        SCREEN_DIM as usize,
                        SCREEN_DIM as usize,
                        RENDER_FPS,
                    )
                    .expect("failed to create render window")
                });

                if !window.is_open() {
                    return Ok(RenderFrame::None);
                }

                window.show(canvas)?;
                Ok(RenderFrame::None)
            }
            RenderMode::RgbArray => {
                let rgb = canvas.pixels_rgb();
                Ok(RenderFrame::RgbArray {
                    width: SCREEN_DIM,
                    height: SCREEN_DIM,
                    data: rgb,
                })
            }
            _ => Ok(RenderFrame::None),
        }
    }
}

impl Env for AcrobotEnv {
    type Obs = Vec<f32>;
    type Act = i64;
    type ObsSpace = BoundedSpace;
    type ActSpace = Discrete;

    fn step(&mut self, action: &i64) -> Result<StepResult<Vec<f32>>> {
        if self.state.is_none() {
            return Err(Error::ResetNeeded { method: "step" });
        }

        if !self.action_space.contains(action) {
            return Err(Error::InvalidAction {
                reason: format!("expected 0, 1, or 2, got {action}"),
            });
        }

        let [t1, t2, dt1, dt2] = self.state.expect("checked above");
        #[allow(clippy::cast_possible_truncation, clippy::cast_sign_loss)]
        // Action already validated.
        let torque = AVAIL_TORQUE[*action as usize];

        // Augmented state for the integrator.
        let s_aug = [t1, t2, dt1, dt2, torque];
        let ns = rk4(&s_aug, DT);

        let new_t1 = wrap(ns[0], -PI, PI);
        let new_t2 = wrap(ns[1], -PI, PI);
        let new_dt1 = bound(ns[2], -MAX_VEL_1, MAX_VEL_1);
        let new_dt2 = bound(ns[3], -MAX_VEL_2, MAX_VEL_2);

        self.state = Some([new_t1, new_t2, new_dt1, new_dt2]);

        let terminated = self.is_terminal();
        let reward = if terminated { 0.0 } else { -1.0 };

        Ok(StepResult {
            obs: self.observation(),
            reward,
            terminated,
            truncated: false,
            info: HashMap::new(),
        })
    }

    fn reset(&mut self, seed: Option<u64>) -> Result<ResetResult<Vec<f32>>> {
        if let Some(s) = seed {
            self.rng = rng::create_rng(Some(s));
        }

        // All four state variables uniform in [-0.1, 0.1].
        let state: [f64; 4] = std::array::from_fn(|_| self.rng.random_range(-0.1..0.1));
        self.state = Some(state);

        Ok(ResetResult {
            obs: self.observation(),
            info: HashMap::new(),
        })
    }

    fn render(&mut self) -> Result<RenderFrame> {
        match self.render_mode {
            RenderMode::None => Ok(RenderFrame::None),
            RenderMode::Ansi => {
                if self.state.is_none() {
                    return Err(Error::ResetNeeded { method: "render" });
                }
                let [t1, t2, dt1, dt2] = self.state.expect("checked above");
                Ok(RenderFrame::Ansi(format!(
                    "Acrobot | θ₁: {t1:+.3} | θ₂: {t2:+.3} | ω₁: {dt1:+.3} | ω₂: {dt2:+.3}"
                )))
            }
            #[cfg(feature = "render")]
            RenderMode::Human | RenderMode::RgbArray => self.render_pixels(),
            #[cfg(not(feature = "render"))]
            _ => Err(Error::UnsupportedRenderMode {
                mode: format!("{:?}", self.render_mode),
            }),
        }
    }

    fn observation_space(&self) -> &BoundedSpace {
        &self.observation_space
    }

    fn action_space(&self) -> &Discrete {
        &self.action_space
    }

    fn render_mode(&self) -> &RenderMode {
        &self.render_mode
    }
}

#[cfg(test)]
mod tests {
    use super::*;

    fn make_env() -> AcrobotEnv {
        AcrobotEnv::new(AcrobotConfig::default()).unwrap()
    }

    #[test]
    fn reset_produces_valid_observation() {
        let mut env = make_env();
        let r = env.reset(Some(42)).unwrap();
        assert_eq!(r.obs.len(), 6);
        assert!(env.observation_space().contains(&r.obs));
    }

    #[test]
    fn step_without_reset_errors() {
        let mut env = make_env();
        assert!(env.step(&0).is_err());
    }

    #[test]
    fn step_invalid_action_errors() {
        let mut env = make_env();
        env.reset(Some(0)).unwrap();
        assert!(env.step(&5).is_err());
    }

    #[test]
    fn step_returns_valid_observation() {
        let mut env = make_env();
        env.reset(Some(42)).unwrap();
        let r = env.step(&2).unwrap();
        assert_eq!(r.obs.len(), 6);
        // cos² + sin² ≈ 1 for both angles.
        let c1 = f64::from(r.obs[1]).mul_add(f64::from(r.obs[1]), f64::from(r.obs[0]).powi(2));
        let c2 = f64::from(r.obs[3]).mul_add(f64::from(r.obs[3]), f64::from(r.obs[2]).powi(2));
        assert!((c1 - 1.0).abs() < 1e-5);
        assert!((c2 - 1.0).abs() < 1e-5);
    }

    #[test]
    fn reward_is_negative_one_before_termination() {
        let mut env = make_env();
        env.reset(Some(42)).unwrap();
        let r = env.step(&1).unwrap();
        if !r.terminated {
            assert!((r.reward - (-1.0)).abs() < f64::EPSILON);
        }
    }

    #[test]
    fn deterministic_with_seed() {
        let mut e1 = make_env();
        let mut e2 = make_env();

        let r1 = e1.reset(Some(99)).unwrap();
        let r2 = e2.reset(Some(99)).unwrap();
        assert_eq!(r1.obs, r2.obs);

        let s1 = e1.step(&2).unwrap();
        let s2 = e2.step(&2).unwrap();
        assert_eq!(s1.obs, s2.obs);
        assert!((s1.reward - s2.reward).abs() < f64::EPSILON);
    }

    #[test]
    fn wrap_angle_works() {
        assert!((wrap(0.0, -PI, PI) - 0.0).abs() < 1e-10);
        assert!((wrap(2.0 * PI, -PI, PI) - 0.0).abs() < 1e-10);
    }
}