Table A1.

DCCT hyperparameters.

Hyperparameter	Value
Observation rendering	(100, 100)
Observation crop	(84, 84)
Hidden units (MLP)	1024
MSA	8
Optimizer	AdamW
$({\beta }_1,{\beta }_2)\to (f_{\theta },{\pi }_{\psi },Q_{\varphi })$	(0.9, 0.999)
$({\beta }_1,{\beta }_2)\to \left(\alpha \right)$	(0.5, 0.999)
Learning rate $(f_{\theta },{\pi }_{\psi },Q_{\varphi })$	2 $\times {10}^{-3}$
Learning rate $\left(\alpha \right)$	2 $\times {10}^{-4}$
Weight decay	3 $\times {10}^{-2}$
Batch Size	512
Convolutional layers	2
Number of filters	36
Transformer encoder layers	6
Convolutional kernal size	3 × 3
Q function EMA $\mathcal{T}$	0.01
Critic target update freq	2
Nonlinearity	ReLU
Encoder EMA $\mathcal{T}$	0.05
Discount $\gamma$	0.99
Initial temperature	0.1