File size: 35,067 Bytes
8a6ca49
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
from typing import List, Tuple, Union

import einops
import numpy as np
import torch
from transformers import AutoModel, PreTrainedModel
from vector_quantize_pytorch import VectorQuantize

from .configuration_actioncodec import ActionCodecConfig
from .modular_actioncodec import PerceiverDecoder, PerceiverEncoder
from .rvq import ResidualVectorQuantize


def trim_trailing_zeros(arr: np.ndarray) -> list[np.ndarray]:
    if arr.shape[0] == 0:
        return []

    b, n = arr.shape

    is_nonzero = arr != 0
    flipped_mask = np.flip(is_nonzero, axis=1)
    last_nonzero_indices = n - 1 - np.argmax(flipped_mask, axis=1)
    any_nonzero_in_row = is_nonzero.any(axis=1)
    new_lengths = (last_nonzero_indices + 1) * any_nonzero_in_row
    result = [arr[i, :length].tolist() for i, length in enumerate(new_lengths)]

    return result


class ActionCodec(PreTrainedModel):
    """ActionCodec: A neural codec for encoding and decoding robot action sequences.

    This model uses a Perceiver-based encoder-decoder architecture with vector quantization
    to convert continuous action sequences into discrete token sequences. It supports
    multiple robot embodiments with different action dimensions and control frequencies.

    The model supports two vector quantization types:
    - VQ (Vector Quantization): Single quantizer
    - RVQ (Residual Vector Quantization): Multiple quantizers for hierarchical encoding

    Key features:
    - Multi-embodiment support: Handle different robots with varying action dimensions
    - Dynamic expansion: Add new robot configurations without retraining
    - Flexible input/output: Support numpy arrays and torch tensors
    """

    config_class = ActionCodecConfig

    def __init__(self, config: ActionCodecConfig):
        """Initialize the ActionCodec model.

        Args:
            config (ActionCodecConfig): Model configuration containing hyperparameters
                and embodiment configurations.

        Raises:
            ValueError: If configuration parameters are invalid.
            NotImplementedError: If the specified VQ type is not supported.
        """
        super().__init__(config)

        # Validate configuration
        if config.n_tokens % config.n_quantizers != 0:
            raise ValueError(f"n_tokens ({config.n_tokens}) must be divisible by n_quantizers ({config.n_quantizers})")

        if config.n_quantizers < 1:
            raise ValueError(f"n_quantizers must be at least 1, got {config.n_quantizers}")

        if config.vq_codebook_size < 1:
            raise ValueError(f"vq_codebook_size must be at least 1, got {config.vq_codebook_size}")

        if config.z_dim < 1:
            raise ValueError(f"z_dim must be at least 1, got {config.z_dim}")

        if not isinstance(config.embodiment_config, dict) or len(config.embodiment_config) == 0:
            raise ValueError(
                "embodiment_config must be a non-empty dictionary mapping embodiment names to configurations"
            )

        self.default_embodiment_id = 0

        # Initialize encoder and decoder
        self.encoder = PerceiverEncoder(config)
        self.decoder = PerceiverDecoder(config)

        # Initialize vector quantizer based on type
        if config.vq_type == "vq":
            if config.n_quantizers != 1:
                raise ValueError(
                    f"VQ type requires n_quantizers=1, got {config.n_quantizers}. Use RVQ type for multiple quantizers."
                )
            self.vq = VectorQuantize(
                dim=config.z_dim,
                codebook_size=config.vq_codebook_size,
                commitment_weight=config.vq_commitment_weight,
                decay=config.vq_decay,
                kmeans_init=config.vq_kmeans_init,
                threshold_ema_dead_code=config.vq_threshold_ema_dead_code,
                rotation_trick=False,
                straight_through=True,
            )
        elif config.vq_type == "rvq":
            if config.n_quantizers < 2:
                raise ValueError(
                    f"RVQ type requires n_quantizers >= 2, got {config.n_quantizers}. Use VQ type for single quantizer."
                )
            self.vq = ResidualVectorQuantize(
                dim=config.z_dim,
                n_codebooks=config.n_quantizers,
                codebook_size=config.vq_codebook_size,
                codebook_dim=config.z_dim,
                quantizer_dropout=config.vq_quantizer_dropout,
                commitment=config.vq_commitment_weight,
            )
        else:
            raise NotImplementedError(f"VQ type '{config.vq_type}' not implemented. Supported types: 'vq', 'rvq'")

        # Store quantization-related attributes
        self.vocab_size = config.vq_codebook_size
        self.num_quantizers = config.n_quantizers
        self.n_tokens_per_quantizer = config.n_tokens // config.n_quantizers

    def expand_embodiment(self, embodiment_config: dict):
        """Dynamically expand the model to support new robot embodiments.

        This method allows adding new robot configurations to the codec without retraining
        the entire model. It updates the encoder and decoder to handle the new action dimensions
        and frequencies while preserving existing functionality for previously configured robots.

        Args:
            embodiment_config (dict): Dictionary mapping embodiment names to their configurations.
                Each configuration should be a dict with keys:
                - "action_dim" (int): Action dimensionality for this embodiment.
                - "freq" (float): Control frequency in Hz.
                - "duration" (float): Default action sequence duration in seconds.
                - "description" (str, optional): Human-readable description.

                Example:
                    {
                        "robot_B": {
                            "action_dim": 10,
                            "freq": 20,
                            "duration": 1.0,
                            "description": "10-dim robot at 20Hz"
                        }
                    }

        Returns:
            ActionCodec: Returns self for method chaining.

        Note:
            - New embodiment keys must not already exist in the current configuration.
            - The model will automatically update max_action_dim if the new embodiment
              has a larger action dimension.
            - Existing embodiments will continue to work with their original configurations.
        """
        if not isinstance(embodiment_config, dict):
            raise TypeError(f"embodiment_config must be a dict, got {type(embodiment_config)}")
        if len(embodiment_config) == 0:
            raise ValueError("embodiment_config cannot be empty")

        # Check for duplicate keys
        overlapping_keys = set(embodiment_config.keys()) & set(self.config.embodiment_config.keys())
        if overlapping_keys:
            raise ValueError(f"The following embodiment keys already exist and cannot be redefined: {overlapping_keys}")

        self.encoder.expand_embodiment(embodiment_config)
        self.decoder.expand_embodiment(embodiment_config)
        self.config.embodiment_config.update(embodiment_config)
        return self

    def _encode(
        self,
        x: torch.Tensor,
        embodiment_ids: torch.Tensor | int | None = None,
        padding_mask: torch.Tensor | None = None,
    ) -> torch.Tensor:
        """Encode action sequences into latent representations.

        Args:
            x (torch.Tensor): Action sequences to encode. Shape: (b, seq_len, max_action_dim).
                Assumes that the action dimension is zero-padded to the max action dimension.
                `seq_len` is supposed to be `int(duration * freq)` for each embodiment and padded to the max sequence length.
            embodiment_ids (torch.Tensor | int): Embodiment IDs. Shape: (b,).
                If int, the same embodiment ID is repeated for all sequences in the batch.
                It specifies the embodiment to encode.
            padding_mask (Optional[torch.Tensor], optional): Padding mask, where `False` values indicate padding. Shape: (b, seq_len). Defaults to None.
                It is used to mask the padding tokens on `seq_len` dimension.

        Returns:
            torch.Tensor: Encoded latent representations. Shape: (b, n_tokens_per_quantizer, z_dim).
        """
        embodiment_ids = embodiment_ids if embodiment_ids is not None else self.default_embodiment_id
        z_e = self.encoder(x, embodiment_ids, padding_mask)
        return z_e

    def _quantize(
        self, z_e: torch.Tensor, return_perplexity: bool = True
    ) -> Tuple[torch.Tensor, torch.Tensor, Union[float, List[float]], torch.Tensor]:
        """Quantize encoded representations using vector quantization.

        Args:
            z_e (torch.Tensor): Encoded latent representations to quantize.
                Shape: (b, n_tokens_per_quantizer, z_dim).
            return_perplexity (bool, optional): Whether to compute and return perplexity.
                Defaults to True.

        Returns:
            Tuple[torch.Tensor, torch.Tensor, Union[float, List[float]], torch.Tensor]:
                A tuple containing:
                - z_q (torch.Tensor): Quantized representations.
                  Shape: (b, n_tokens_per_quantizer, z_dim).
                - indices (torch.Tensor): Quantization indices.
                  Shape: (b, n_tokens_per_quantizer) for VQ or (b, n_tokens_per_quantizer, n_quantizers) for RVQ.
                - perplexity (Union[float, List[float]]): Codebook perplexity.
                  Float for single quantizer, List[float] for multiple quantizers.
                - commit_loss (torch.Tensor): Commitment loss scalar tensor.
        """
        if isinstance(self.vq, ResidualVectorQuantize):
            z_q, indices, _, commitment_loss, codebook_loss = self.vq(z_e)
            commit_loss = commitment_loss.mean() + codebook_loss.mean()
        elif isinstance(self.vq, VectorQuantize):
            z_q, indices, commit_loss = self.vq(z_e)
        else:
            raise NotImplementedError(f"VQ type {type(self.vq)} not implemented")

        if return_perplexity:
            if len(indices.size()) < 3:
                indices = indices.unsqueeze(-1)
            perplexity = []
            for k in range(indices.size(-1)):
                this_indices = indices[:, :, k]
                indices_count = torch.bincount(this_indices.view(-1), minlength=self.vq.codebook_size)
                if torch.distributed.is_initialized() and torch.distributed.get_world_size() > 1:
                    torch.distributed.all_reduce(indices_count)
                this_avg_probs = indices_count.float() / indices_count.sum()
                perplexity.append(((-(this_avg_probs * torch.log(this_avg_probs + 1e-10)).sum()).exp().item()))
        else:
            perplexity = 0

        return z_q, indices, perplexity, commit_loss

    def _dequantize(self, indices: torch.Tensor) -> torch.Tensor:
        """Dequantize token indices back to continuous latent representations.

        Args:
            indices (torch.Tensor): Quantization indices. Shape depends on quantizer type:
                - For VQ: (b, n_tokens) or (b, n_tokens, 1)
                - For RVQ: (b, n_tokens_per_quantizer, n_quantizers)

        Returns:
            torch.Tensor: Dequantized latent representations.
                Shape: (b, n_tokens_per_quantizer, z_dim)
        """
        if self.num_quantizers == 1:
            if len(indices.size()) == 3:
                indices = indices.squeeze(-1)
        if isinstance(self.vq, ResidualVectorQuantize):
            z_q = self.vq.from_codes(indices)[0]
        elif isinstance(self.vq, VectorQuantize):
            z_q = self.vq.get_output_from_indices(indices)
        else:
            raise NotImplementedError(f"VQ type {type(self.vq)} not implemented in _dequantize")
        return z_q

    def _decode(
        self, z_q: torch.Tensor, embodiment_ids: torch.Tensor | int | None = None, durations: torch.Tensor | None = None
    ) -> Tuple[torch.Tensor, torch.Tensor]:
        """Decode quantized latent representations into action sequences.

        Args:
            z_q (torch.Tensor): Quantized latent representations.
                Shape: (b, n_tokens_per_quantizer, z_dim).
            embodiment_ids (Union[torch.Tensor, int, None], optional): Embodiment IDs.
                Shape: (b,) if tensor. If int, the same embodiment ID is used for all
                sequences. Defaults to None, which uses `self.default_embodiment_id`.
            durations (torch.Tensor | None, optional): Duration of each action sequence in seconds.
                Shape: (b,). If None, uses default duration from embodiment_config.
                Defaults to None.

        Returns:
            Tuple[torch.Tensor, torch.Tensor]: A tuple containing:
                - x_recon (torch.Tensor): Reconstructed action sequences.
                  Shape: (b, seq_len, max_action_dim).
                - padding_mask (torch.Tensor): Padding mask indicating valid timesteps.
                  Shape: (b, seq_len), where True indicates valid timesteps.
        """
        embodiment_ids = embodiment_ids if embodiment_ids is not None else self.default_embodiment_id
        x_recon, padding_mask = self.decoder(z_q, embodiment_ids, durations)
        return x_recon, padding_mask

    @torch.no_grad()
    def encode(
        self,
        x: Union[np.ndarray, torch.Tensor],
        embodiment_ids: Union[List[int], int, None] = None,
        padding_mask: Union[List[bool], np.ndarray, torch.Tensor, None] = None,
        **kwargs,
    ) -> List[List[int]]:
        """Encode action sequences into latent representations (token indices).

        This method converts action sequences into discrete token indices using the encoder
        and vector quantizer. The input can be either a numpy array or torch tensor.

        Args:
            x (Union[np.ndarray, torch.Tensor]): Action sequences to encode.
                Shape: (b, seq_len, max_action_dim).
                Assumes that the action dimension is zero-padded to the max action dimension.
                `seq_len` is supposed to be `int(duration * freq)` for each embodiment and
                padded to the max sequence length.
            embodiment_ids (Union[List[int], int, None], optional): Embodiment IDs.
                Shape: (b,) if list. If int, the same embodiment ID is repeated for all
                sequences in the batch. It specifies the embodiment to encode.
                Defaults to None, which uses `self.default_embodiment_id`.
            padding_mask (Union[List[bool], np.ndarray, torch.Tensor, None], optional):
                Padding mask, where `False` values indicate padding. Shape: (b, seq_len).
                Defaults to None. It is used to mask the padding tokens on `seq_len` dimension.
            **kwargs: Additional keyword arguments (currently unused, reserved for future use).

        Returns:
            List[List[int]]: List of token sequences. Shape: (b, n_tokens), where n_tokens
            is determined by the model configuration (typically `config.n_tokens`).

        Raises:
            ValueError: If input shapes are invalid or incompatible with the model configuration.
            TypeError: If input types are not supported.

        Examples:
            >>> import numpy as np
            >>> # Using numpy array
            >>> x = np.random.randn(2, 10, 7).astype(np.float32)
            >>> tokens = model.encode(x, embodiment_ids=[0, 0])
            >>> # Using torch tensor
            >>> x_tensor = torch.randn(2, 10, 7)
            >>> tokens = model.encode(x_tensor, embodiment_ids=[0, 0])
        """
        self.eval()

        # Validate and convert input x
        if isinstance(x, np.ndarray):
            if x.ndim != 3:
                raise ValueError(
                    f"Expected 3D input array (batch, seq_len, action_dim), got {x.ndim}D array with shape {x.shape}"
                )
            x_tensor = torch.tensor(x, dtype=self.dtype, device=self.device)
        elif isinstance(x, torch.Tensor):
            if x.ndim != 3:
                raise ValueError(
                    f"Expected 3D tensor (batch, seq_len, action_dim), got {x.ndim}D tensor with shape {x.shape}"
                )
            x_tensor = x.to(dtype=self.dtype, device=self.device)
        else:
            raise TypeError(f"Input x must be numpy.ndarray or torch.Tensor, got {type(x)}")

        # Validate batch size
        batch_size = x_tensor.shape[0]
        if batch_size == 0:
            raise ValueError("Batch size must be at least 1")

        # Handle embodiment_ids
        embodiment_ids = embodiment_ids if embodiment_ids is not None else self.default_embodiment_id
        if isinstance(embodiment_ids, int):
            if not 0 <= embodiment_ids < len(self.config.embodiment_config):
                raise ValueError(
                    f"embodiment_id {embodiment_ids} is out of range [0, {len(self.config.embodiment_config)}). "
                    f"Available embodiment IDs: {list(range(len(self.config.embodiment_config)))}"
                )
            embodiment_ids_tensor = torch.tensor([embodiment_ids] * batch_size, dtype=torch.long, device=self.device)
        elif isinstance(embodiment_ids, list):
            if len(embodiment_ids) != batch_size:
                raise ValueError(
                    f"Length of embodiment_ids ({len(embodiment_ids)}) must match batch size ({batch_size})"
                )
            for eid in embodiment_ids:
                if not isinstance(eid, int) or not 0 <= eid < len(self.config.embodiment_config):
                    raise ValueError(
                        f"Invalid embodiment_id {eid}. Must be an integer in range [0, {len(self.config.embodiment_config)})"
                    )
            embodiment_ids_tensor = torch.tensor(embodiment_ids, dtype=torch.long, device=self.device)
        else:
            raise TypeError(f"embodiment_ids must be int, List[int], or None, got {type(embodiment_ids)}")

        # Handle padding_mask
        padding_mask_tensor = None
        if padding_mask is not None:
            if isinstance(padding_mask, (list, np.ndarray)):
                padding_mask_tensor = torch.tensor(padding_mask, dtype=torch.bool, device=self.device)
            elif isinstance(padding_mask, torch.Tensor):
                padding_mask_tensor = padding_mask.to(dtype=torch.bool, device=self.device)
            else:
                raise TypeError(
                    f"padding_mask must be List[bool], np.ndarray, torch.Tensor, or None, got {type(padding_mask)}"
                )
            if padding_mask_tensor.shape != (batch_size, x_tensor.shape[1]):
                raise ValueError(
                    f"padding_mask shape {padding_mask_tensor.shape} does not match expected shape "
                    f"({batch_size}, {x_tensor.shape[1]})"
                )

        with torch.no_grad():
            z_e = self._encode(x_tensor, embodiment_ids_tensor, padding_mask_tensor)
            _, indices, _, _ = self._quantize(z_e, return_perplexity=False)

            # Reshape indices: for RVQ, indices shape is (b, n, s), for VQ it's (b, n)
            if len(indices.size()) > 2:
                codes_list = einops.rearrange(indices, "b n s -> b (s n)").cpu()
            else:
                codes_list = indices.cpu()

            codes_list = codes_list.tolist()
            return codes_list

    @torch.no_grad()
    def decode(
        self,
        tokens: Union[List[List[int]], np.ndarray, torch.Tensor],
        embodiment_ids: Union[List[int], int, None] = None,
        durations: Union[List[float], np.ndarray, torch.Tensor, None] = None,
        **kwargs,
    ) -> Tuple[np.ndarray, np.ndarray]:
        """Decode token sequences into action sequences.

        This method reconstructs action sequences from discrete token indices using the
        vector quantizer and decoder. The input tokens can be a list of lists, numpy array,
        or torch tensor.

        Args:
            tokens (Union[List[List[int]], np.ndarray, torch.Tensor]): Token sequences to decode.
                Shape: (b, n_tokens), where n_tokens must be divisible by `n_tokens_per_quantizer`.
                For RVQ, tokens are interleaved: [q0_t0, q1_t0, ..., qN_t0, q0_t1, ...].
            embodiment_ids (Union[List[int], int, None], optional): Embodiment IDs.
                Shape: (b,) if list. If int, the same embodiment ID is repeated for all
                sequences in the batch. It specifies the embodiment to decode.
                Defaults to None, which uses `self.default_embodiment_id`.
            durations (Union[List[float], np.ndarray, torch.Tensor, None], optional):
                Duration of each action sequence in seconds. Shape: (b,).
                If None, the duration is inferred from the default values in `embodiment_config`.
                Defaults to None.
            **kwargs: Additional keyword arguments (currently unused, reserved for future use).

        Returns:
            Tuple[np.ndarray, np.ndarray]: A tuple containing:
                - reconstructed_actions: Reconstructed action sequences.
                  Shape: (b, seq_len, max_action_dim).
                - padding_mask: Padding mask indicating valid timesteps.
                  Shape: (b, seq_len), where True indicates valid timesteps.

        Raises:
            ValueError: If token sequence length is invalid or incompatible with the model configuration.
            TypeError: If input types are not supported.

        Examples:
            >>> # Using list of lists
            >>> tokens = [[1, 2, 3, 4, 5, 6, 7, 8], [9, 10, 11, 12, 13, 14, 15, 16]]
            >>> actions, mask = model.decode(tokens, embodiment_ids=[0, 0])
            >>> # Using numpy array
            >>> tokens_np = np.array([[1, 2, 3, 4], [5, 6, 7, 8]])
            >>> actions, mask = model.decode(tokens_np, embodiment_ids=[0, 0])
            >>> # Using torch tensor
            >>> tokens_tensor = torch.tensor([[1, 2, 3, 4], [5, 6, 7, 8]])
            >>> actions, mask = model.decode(tokens_tensor, embodiment_ids=[0, 0])
        """
        self.eval()

        # Validate and convert input tokens
        if isinstance(tokens, list):
            if not all(isinstance(seq, list) for seq in tokens):
                raise TypeError("If tokens is a list, all elements must be lists")
            if len(tokens) == 0:
                raise ValueError("Tokens list cannot be empty")
            if not all(isinstance(val, (int, np.integer)) for seq in tokens for val in seq):
                raise TypeError("All token values must be integers")
            tokens_tensor = torch.tensor(tokens, dtype=torch.long, device=self.device)
        elif isinstance(tokens, np.ndarray):
            if tokens.ndim != 2:
                raise ValueError(
                    f"Expected 2D array (batch, n_tokens), got {tokens.ndim}D array with shape {tokens.shape}"
                )
            if not np.issubdtype(tokens.dtype, np.integer):
                raise TypeError(f"Tokens array must have integer dtype, got {tokens.dtype}")
            tokens_tensor = torch.tensor(tokens, dtype=torch.long, device=self.device)
        elif isinstance(tokens, torch.Tensor):
            if tokens.ndim != 2:
                raise ValueError(
                    f"Expected 2D tensor (batch, n_tokens), got {tokens.ndim}D tensor with shape {tokens.shape}"
                )
            if not tokens.dtype.is_integer:
                raise TypeError(f"Tokens tensor must have integer dtype, got {tokens.dtype}")
            tokens_tensor = tokens.to(dtype=torch.long, device=self.device)
        else:
            raise TypeError(f"tokens must be List[List[int]], np.ndarray, or torch.Tensor, got {type(tokens)}")

        batch_size, n_tokens = tokens_tensor.shape
        if batch_size == 0:
            raise ValueError("Batch size must be at least 1")
        if n_tokens == 0:
            raise ValueError("Token sequence length must be at least 1")

        # Validate token sequence length
        if n_tokens % self.n_tokens_per_quantizer != 0:
            raise ValueError(
                f"Token sequence length ({n_tokens}) must be divisible by tokens per quantizer "
                f"({self.n_tokens_per_quantizer}). Total tokens: {n_tokens}, "
                f"Expected multiple of: {self.n_tokens_per_quantizer}. "
                f"Number of quantizers: {self.num_quantizers}, Total tokens per sequence: {self.config.n_tokens}"
            )

        # Validate token values are within codebook range
        if tokens_tensor.min() < 0 or tokens_tensor.max() >= self.vocab_size:
            raise ValueError(
                f"Token values must be in range [0, {self.vocab_size}), "
                f"got range [{tokens_tensor.min().item()}, {tokens_tensor.max().item()}]"
            )

        # Handle embodiment_ids
        embodiment_ids = embodiment_ids if embodiment_ids is not None else self.default_embodiment_id
        if isinstance(embodiment_ids, int):
            if not 0 <= embodiment_ids < len(self.config.embodiment_config):
                raise ValueError(
                    f"embodiment_id {embodiment_ids} is out of range [0, {len(self.config.embodiment_config)}). "
                    f"Available embodiment IDs: {list(range(len(self.config.embodiment_config)))}"
                )
            embodiment_ids_tensor = torch.tensor([embodiment_ids] * batch_size, dtype=torch.long, device=self.device)
        elif isinstance(embodiment_ids, list):
            if len(embodiment_ids) != batch_size:
                raise ValueError(
                    f"Length of embodiment_ids ({len(embodiment_ids)}) must match batch size ({batch_size})"
                )
            for eid in embodiment_ids:
                if not isinstance(eid, int) or not 0 <= eid < len(self.config.embodiment_config):
                    raise ValueError(
                        f"Invalid embodiment_id {eid}. Must be an integer in range [0, {len(self.config.embodiment_config)})"
                    )
            embodiment_ids_tensor = torch.tensor(embodiment_ids, dtype=torch.long, device=self.device)
        else:
            raise TypeError(f"embodiment_ids must be int, List[int], or None, got {type(embodiment_ids)}")

        # Handle durations
        durations_tensor = None
        if durations is not None:
            if isinstance(durations, (list, np.ndarray)):
                durations_tensor = torch.tensor(durations, dtype=torch.float32, device=self.device)
            elif isinstance(durations, torch.Tensor):
                durations_tensor = durations.to(dtype=torch.float32, device=self.device)
            else:
                raise TypeError(
                    f"durations must be List[float], np.ndarray, torch.Tensor, or None, got {type(durations)}"
                )
            if durations_tensor.ndim != 1:
                raise ValueError(
                    f"durations must be 1D, got {durations_tensor.ndim}D with shape {durations_tensor.shape}"
                )
            if len(durations_tensor) != batch_size:
                raise ValueError(f"Length of durations ({len(durations_tensor)}) must match batch size ({batch_size})")
            if (durations_tensor <= 0).any():
                raise ValueError("All durations must be positive")

        # Reshape tokens for dequantization: (b, n_tokens) -> (b, n_tokens_per_quantizer, n_quantizers)
        indices = einops.rearrange(tokens_tensor, "b (n m) -> b m n", m=self.n_tokens_per_quantizer)

        with torch.no_grad():
            z_q = self._dequantize(indices)
            x_recon, padding_mask = self._decode(z_q, embodiment_ids_tensor, durations_tensor)

        return x_recon.float().cpu().numpy(), padding_mask.float().cpu().numpy()

    def forward(
        self,
        x: Union[torch.Tensor, np.ndarray],
        embodiment_ids: Union[torch.Tensor, int, List[int], None] = None,
        padding_mask: Union[torch.Tensor, List[bool], np.ndarray, None] = None,
    ) -> Tuple[torch.Tensor, torch.Tensor]:
        """Forward pass through the full ActionCodec pipeline.

        This method performs encoding, quantization, and decoding in a single forward pass.
        It is primarily used during training to compute reconstruction loss and commitment loss.
        Both numpy arrays and torch tensors are supported as input.

        Args:
            x (Union[torch.Tensor, np.ndarray]): Action sequences to process.
                Shape: (b, seq_len, max_action_dim).
            embodiment_ids (Union[torch.Tensor, int, List[int], None], optional):
                Embodiment IDs. Shape: (b,) if tensor or list. If int, same ID for all sequences.
                Defaults to None, which uses `self.default_embodiment_id`.
            padding_mask (Union[torch.Tensor, List[bool], np.ndarray, None], optional):
                Padding mask. Shape: (b, seq_len). Defaults to None.

        Returns:
            Tuple[torch.Tensor, torch.Tensor]: A tuple containing:
                - x_recon (torch.Tensor): Reconstructed action sequences.
                  Shape: (b, seq_len, max_action_dim).
                - recon_mask (torch.Tensor): Reconstruction mask indicating valid timesteps.
                  Shape: (b, seq_len), where True indicates valid timesteps.

        Note:
            - For inference use cases, prefer using `encode()` and `decode()` methods separately.
            - If you need token indices, use the `encode()` method instead.
        """
        # Convert numpy array to torch tensor if needed
        if isinstance(x, np.ndarray):
            x = torch.tensor(x, dtype=self.dtype, device=self.device)

        # Handle embodiment_ids conversion
        if isinstance(embodiment_ids, list):
            embodiment_ids = torch.tensor(embodiment_ids, device=x.device, dtype=torch.long)
        elif isinstance(embodiment_ids, int):
            # Keep as int, will be handled by _encode
            pass

        # Handle padding_mask conversion
        if isinstance(padding_mask, (list, np.ndarray)):
            padding_mask = torch.tensor(padding_mask, device=x.device, dtype=torch.bool)

        # Full forward pass: encode -> quantize -> decode
        z_e = self._encode(x, embodiment_ids, padding_mask)
        z_q, indices, perplexity, commit_loss = self._quantize(z_e, return_perplexity=True)
        x_recon, recon_mask = self._decode(z_q, embodiment_ids)

        return x_recon, recon_mask


AutoModel.register(ActionCodecConfig, ActionCodec)

__all__ = ["ActionCodec"]


if __name__ == "__main__":
    print("=== ActionCodec Comprehensive Test ===\n")

    # 1. Configuration Setup (RVQ enabled with n_quantizers=4)
    initial_config = {
        "robot_A": {"action_dim": 7, "freq": 10, "duration": 1, "description": "Robot A"},
    }

    # We set n_quantizers=4 to test Residual VQ logic
    config = ActionCodecConfig(
        embodiment_config=initial_config,
        n_tokens=16,  # Total tokens per sequence (latent_len * n_quantizers)
        n_quantizers=4,  # RVQ depth
        vq_type="rvq",
        vq_codebook_size=256,
        encoder_dim=128,
        decoder_dim=128,
    )

    # Expected latent sequence length = n_tokens / n_quantizers = 16 / 4 = 4
    latent_seq_len = int(config.n_tokens // config.n_quantizers)
    print(f"Config: {config.n_quantizers} quantizers, {latent_seq_len} latent vectors per sequence.")

    codec = ActionCodec(config)
    codec.eval()

    # 2. Basic Encode/Decode Test
    print("\n--- Test 1: Basic Encode/Decode ---")
    batch_size = 2
    seq_len_A = 10  # 10Hz * 1s

    # Create random action data for Robot A (ID 0)
    x = np.random.randn(batch_size, seq_len_A, 7).astype(np.float32)
    # Masking: Second item in batch is half padding
    padding_mask = np.ones((batch_size, seq_len_A), dtype=bool)
    padding_mask[1, 5:] = False

    embodiment_ids = [0, 0]

    # Encode
    codes = codec.encode(x, embodiment_ids, padding_mask)
    print(f"Encoded codes shape (list length): {len(codes)} x {len(codes[0])}")

    # Validate code length
    assert len(codes[0]) == config.n_tokens, f"Expected {config.n_tokens} tokens, got {len(codes[0])}"

    # Decode
    x_recon, recon_mask = codec.decode(codes, embodiment_ids)
    print(f"Reconstructed shape: {x_recon.shape}")
    print(f"Recon mask shape: {recon_mask.shape}")

    assert x_recon.shape == (batch_size, seq_len_A, 7)  # Should imply zero-padding to max dim 7

    # 3. Expansion Test
    print("\n--- Test 2: Dynamic Expansion ---")
    new_robot_config = {"robot_B": {"action_dim": 10, "freq": 20, "duration": 1, "description": "Robot B (Larger)"}}

    print("Expanding codec to include Robot B (10 dims, 20Hz)...")
    codec.expand_embodiment(new_robot_config)

    assert codec.encoder.max_action_dim == 10
    assert codec.decoder.max_action_dim == 10
    print("✅ Expansion successful.")

    # 4. Mixed Batch Test (Old + New Robot)
    print("\n--- Test 3: Mixed Batch Inference ---")

    # Batch: [Robot A, Robot B]
    # Robot A: 10Hz, 1s -> 10 steps. Dims 7.
    # Robot B: 20Hz, 1s -> 20 steps. Dims 10.
    # Batch Max Steps: 20. Batch Max Dims: 10.

    batch_x_mixed = np.zeros((2, 20, 10), dtype=np.float32)

    # Fill Robot A data (index 0)
    data_A = np.random.randn(10, 7)
    batch_x_mixed[0, :10, :7] = data_A

    # Fill Robot B data (index 1)
    data_B = np.random.randn(20, 10)
    batch_x_mixed[1, :20, :10] = data_B

    # Embodiment IDs: 0 for A, 1 for B
    # Note: expand_embodiment appends. Original was 0, new is 1.
    mixed_ids = [0, 1]

    # Encode Mask
    mixed_mask = np.zeros((2, 20), dtype=bool)
    mixed_mask[0, :10] = True
    mixed_mask[1, :20] = True

    print("Encoding mixed batch...")
    mixed_codes = codec.encode(batch_x_mixed, mixed_ids, mixed_mask)

    print("Decoding mixed batch...")
    # Explicit durations (optional, but good for verification if we wanted to override defaults)
    durations = [1, 1]
    x_recon_mixed, dec_mask_mixed = codec.decode(mixed_codes, mixed_ids, durations)

    print(f"Mixed Recon Shape: {x_recon_mixed.shape}")

    # Validation
    # Robot A output check (mask should be True for first 10, False for rest)
    valid_A = dec_mask_mixed[0].sum()
    valid_B = dec_mask_mixed[1].sum()

    print(f"Valid steps detected by Decoder: Robot A={valid_A}, Robot B={valid_B}")

    assert valid_A == 10
    assert valid_B == 20

    # Check dimensionality preservation
    # Robot A's reconstruction in dims 7-9 should be noise or zero (depending on implementation),
    # but dims 0-6 should contain signal.
    print("✅ Mixed batch processed successfully.")

    print("\n✨ All systems go.")