CrafterVPT: Behavioral Foundation Model

• The paper introduces CrafterVPT as a robust behavioral prior that serves as a low-level controller in the lightweight Crafter environment.
• It employs behavior cloning on the synthetic CrafterPlay dataset with strategic no-op filtering to mitigate action imbalance and enhance performance.
• Evaluation on the Crafter benchmark shows that CrafterVPT outperforms previous baselines while requiring additional planning layers for multi-step tasks.

Searching arXiv for the cited CrafterDojo paper and the original VPT/Crafter context papers. CrafterVPT, also written C-VPT, is the behavioral foundation model in the broader CrafterDojo suite, introduced to provide a strong unconditional policy prior for the Crafter environment (Park et al., 19 Aug 2025). Within that suite, CrafterVPT serves as the behavioral prior, CrafterCLIP as the vision-language grounding component, and CrafterSteve-1 as the instruction-following component. The model is positioned as a Crafter analogue of behavioral pretraining in Minecraft: rather than requiring higher-level agents to learn low-level motor behavior from scratch, it supplies a pretrained low-level controller that can be reused for instruction following, hierarchical planning, and reinforcement-learning fine-tuning (Park et al., 19 Aug 2025).

1. Conceptual role within CrafterDojo

CrafterVPT was introduced because Crafter, despite being a lightweight, open-source, Minecraft-like environment, had lacked the kind of foundation-model ecosystem that had supported progress in Minecraft research (Park et al., 19 Aug 2025). The immediate motivation was threefold: to provide a behavioral prior for Crafter agents, to enable hierarchical agents in which a planner issues high-level instructions and a CrafterVPT-like controller handles primitive actions, and to unlock prototyping-friendly embodied agent research in a lightweight environment that avoids the engineering overhead of Minecraft (Park et al., 19 Aug 2025).

This placement makes CrafterVPT more than a standalone imitation-learned policy. In the architecture of CrafterDojo, it is the initial behavioral layer on which later components depend. CrafterCLIP is trained on captions derived from the same CrafterPlay trajectories, and CrafterSteve-1 is constructed by conditioning the CrafterVPT backbone on goal embeddings from CrafterCLIP, thereby converting the unconditional policy into an instruction-following agent (Park et al., 19 Aug 2025). The paper therefore treats CrafterVPT as the reusable low-level “motor system” of the suite.

A plausible implication is that CrafterVPT occupies the same systems role in Crafter that behavioral priors occupy in other hierarchical embodied-agent stacks: it amortizes low-level control and allows downstream modules to operate at a more semantic level. The paper itself stops short of presenting it as sufficient for planning-heavy tasks on its own (Park et al., 19 Aug 2025).

2. Dataset construction and training regime

CrafterVPT is trained by behavior cloning / imitation learning on a synthetic expert dataset called CrafterPlay (Park et al., 19 Aug 2025). Because Crafter lacks large online human gameplay corpora, the dataset was built using an Expert Behavior Generator toolkit. The construction pipeline is explicitly described as: training an expert policy in Craftax-Classic-Symbolic using PPO-RNN, rolling out that expert to generate trajectories, and then using those trajectories as demonstration data for CrafterVPT (Park et al., 19 Aug 2025).

The expert policy used for data generation was trained for 10B timesteps in Craftax-Classic-Symbolic, following Craftax’s PPO-RNN protocol, with an added survival penalty to avoid collapse into poor survival behavior after all achievements were already unlocked (Park et al., 19 Aug 2025). The resulting CrafterPlay dataset contains 20,000 episodes, about 180M timesteps in the appendix, and an average length of approximately 9,012 steps per episode. Its trajectories are written as

$\tau_i = \{(o_t,a_t)\}_{t=1}^{T},$

that is, observation-action trajectories with explicit action labels (Park et al., 19 Aug 2025).

The availability of ground-truth actions is a central distinction from the original Minecraft VPT setup. CrafterVPT does not require an Inverse Dynamics Model (IDM) to infer actions from unlabeled video, because the synthetic CrafterPlay pipeline already provides action supervision (Park et al., 19 Aug 2025). This alters both the data-engineering pathway and the training objective: the system is not trying to reconstruct latent actions from videos, but directly fit a policy to labeled trajectories.

The paper also reports scaling observations for the upstream expert and the downstream imitation learner. Expert-policy performance continued improving until around 10B timesteps, which justified using that endpoint for dataset generation. On the imitation side, training on larger CrafterPlay subsets improved performance and converged around 18,000 episodes (Park et al., 19 Aug 2025). This suggests that CrafterVPT’s effectiveness is tied not only to architecture but also to the scale and quality of its synthetic demonstration corpus.

3. Model architecture and optimization objective

CrafterVPT uses the identical architecture as VPT:

$$\begin{aligned} \text{Image Encoder: } \hspace{0.5em} && x_t & = \text{ResNet}_\theta (o_t) \ \text{TransformerXL: } \hspace{0.5em} && \tilde{x}_{1:t} & = \mathrm{TrXL}_\theta(x_{1:t}) \ \text{Policy Head: } \hspace{0.5em} && a_t & \sim \pi_\theta (a_t \mid \tilde{x}_{1:t}). \end{aligned}$$

Here, $o_t$ is the current observation frame, $x_t$ is the image embedding produced by a ResNet, $\tilde{x}_{1:t}$ is the contextualized history representation from Transformer-XL, $a_t$ is the sampled action, and $\pi_\theta$ is the policy parameterized by $\theta$ (Park et al., 19 Aug 2025). In operational terms, this is a pixel-based recurrent policy combining an image encoder, a sequence model, and an action predictor.

Three model sizes are trained.

Variant	Params	Recurrence layers
tiny	3.2M	2
base	15.9M	4
large	64.3M	4

The associated hidden dimensions are reported as follows: tiny uses $d_\text{model}=128$, $d_\text{ff}=512$, and 4 heads; base uses $$\begin{aligned} \text{Image Encoder: } \hspace{0.5em} && x_t & = \text{ResNet}_\theta (o_t) \ \text{TransformerXL: } \hspace{0.5em} && \tilde{x}_{1:t} & = \mathrm{TrXL}_\theta(x_{1:t}) \ \text{Policy Head: } \hspace{0.5em} && a_t & \sim \pi_\theta (a_t \mid \tilde{x}_{1:t}). \end{aligned}$$0, $$\begin{aligned} \text{Image Encoder: } \hspace{0.5em} && x_t & = \text{ResNet}_\theta (o_t) \ \text{TransformerXL: } \hspace{0.5em} && \tilde{x}_{1:t} & = \mathrm{TrXL}_\theta(x_{1:t}) \ \text{Policy Head: } \hspace{0.5em} && a_t & \sim \pi_\theta (a_t \mid \tilde{x}_{1:t}). \end{aligned}$$1, and 4 heads; large uses $$\begin{aligned} \text{Image Encoder: } \hspace{0.5em} && x_t & = \text{ResNet}_\theta (o_t) \ \text{TransformerXL: } \hspace{0.5em} && \tilde{x}_{1:t} & = \mathrm{TrXL}_\theta(x_{1:t}) \ \text{Policy Head: } \hspace{0.5em} && a_t & \sim \pi_\theta (a_t \mid \tilde{x}_{1:t}). \end{aligned}$$2, $$\begin{aligned} \text{Image Encoder: } \hspace{0.5em} && x_t & = \text{ResNet}_\theta (o_t) \ \text{TransformerXL: } \hspace{0.5em} && \tilde{x}_{1:t} & = \mathrm{TrXL}_\theta(x_{1:t}) \ \text{Policy Head: } \hspace{0.5em} && a_t & \sim \pi_\theta (a_t \mid \tilde{x}_{1:t}). \end{aligned}$$3, and 8 heads (Park et al., 19 Aug 2025). The paper notes that the largest CrafterVPT corresponds roughly to the smallest VPT variant in the Minecraft literature (Park et al., 19 Aug 2025).

The training objective is standard behavior cloning:

$$\begin{aligned} \text{Image Encoder: } \hspace{0.5em} && x_t & = \text{ResNet}_\theta (o_t) \ \text{TransformerXL: } \hspace{0.5em} && \tilde{x}_{1:t} & = \mathrm{TrXL}_\theta(x_{1:t}) \ \text{Policy Head: } \hspace{0.5em} && a_t & \sim \pi_\theta (a_t \mid \tilde{x}_{1:t}). \end{aligned}$$4

In this expression, $$\begin{aligned} \text{Image Encoder: } \hspace{0.5em} && x_t & = \text{ResNet}_\theta (o_t) \ \text{TransformerXL: } \hspace{0.5em} && \tilde{x}_{1:t} & = \mathrm{TrXL}_\theta(x_{1:t}) \ \text{Policy Head: } \hspace{0.5em} && a_t & \sim \pi_\theta (a_t \mid \tilde{x}_{1:t}). \end{aligned}$$5 is the CrafterPlay dataset, $$\begin{aligned} \text{Image Encoder: } \hspace{0.5em} && x_t & = \text{ResNet}_\theta (o_t) \ \text{TransformerXL: } \hspace{0.5em} && \tilde{x}_{1:t} & = \mathrm{TrXL}_\theta(x_{1:t}) \ \text{Policy Head: } \hspace{0.5em} && a_t & \sim \pi_\theta (a_t \mid \tilde{x}_{1:t}). \end{aligned}$$6 is the observation history up to time $$\begin{aligned} \text{Image Encoder: } \hspace{0.5em} && x_t & = \text{ResNet}_\theta (o_t) \ \text{TransformerXL: } \hspace{0.5em} && \tilde{x}_{1:t} & = \mathrm{TrXL}_\theta(x_{1:t}) \ \text{Policy Head: } \hspace{0.5em} && a_t & \sim \pi_\theta (a_t \mid \tilde{x}_{1:t}). \end{aligned}$$7, and $$\begin{aligned} \text{Image Encoder: } \hspace{0.5em} && x_t & = \text{ResNet}_\theta (o_t) \ \text{TransformerXL: } \hspace{0.5em} && \tilde{x}_{1:t} & = \mathrm{TrXL}_\theta(x_{1:t}) \ \text{Policy Head: } \hspace{0.5em} && a_t & \sim \pi_\theta (a_t \mid \tilde{x}_{1:t}). \end{aligned}$$8 is the negative log-likelihood of the expert action under the policy (Park et al., 19 Aug 2025). The objective therefore optimizes next-action likelihood under recurrent pixel-conditioned history, rather than reward maximization directly.

4. Preprocessing, no-op imbalance, and ablations

A major practical issue in CrafterVPT training is noop filtering. CrafterPlay contained about 60% no-op actions, which the authors identify as problematic because excessive inactive steps degrade behavior cloning, while eliminating all no-ops would remove useful waiting behaviors (Park et al., 19 Aug 2025). The adopted scheme is selective: it removes noop sequences shorter than 20 steps and preserves longer no-op stretches, since longer idle periods can encode meaningful behaviors such as waiting for safety (Park et al., 19 Aug 2025).

This preprocessing step reduces the no-op ratio to 4.6% and improves CrafterVPT performance (Park et al., 19 Aug 2025). The appendix further quantifies the structure of the imbalance: 60% of all recorded actions are no-op, 40.6% of no-op actions occur during sleep periods, and sleep-related no-ops account for 24% of all recorded actions (Park et al., 19 Aug 2025). These numbers matter because they show that the issue is not merely annotation redundancy; a substantial fraction of idle behavior is behaviorally meaningful.

The ablation is one of the clearest empirical findings in the paper. If CrafterVPT is trained without this filtering, its Crafter Score drops to 38.1, about a 38% decrease from the final model (Park et al., 19 Aug 2025). The paper explicitly interprets the chosen filtering strategy as a compromise between suppressing pathological class imbalance and preserving strategic waiting.

Other ablations reinforce the scaling picture. Performance improves as the expert policy used to generate CrafterPlay is trained longer, up to around 10B timesteps, and it improves with more imitation data, converging around 18,000 episodes (Park et al., 19 Aug 2025). Taken together, these results indicate that the behavioral prior is sensitive to dataset composition, expert competence, and action-frequency skew, not only to raw model size.

5. Evaluation protocol and quantitative performance

CrafterVPT is evaluated in the standard Crafter benchmark using two metrics: Crafter Score, defined as the geometric mean of success across the 22 achievements and scaled to $$\begin{aligned} \text{Image Encoder: } \hspace{0.5em} && x_t & = \text{ResNet}_\theta (o_t) \ \text{TransformerXL: } \hspace{0.5em} && \tilde{x}_{1:t} & = \mathrm{TrXL}_\theta(x_{1:t}) \ \text{Policy Head: } \hspace{0.5em} && a_t & \sim \pi_\theta (a_t \mid \tilde{x}_{1:t}). \end{aligned}$$9, and Return, defined as normalized cumulative reward (Park et al., 19 Aug 2025). The appendix gives the Crafter Score formula:

$o_t$0

where $o_t$1 is the success rate for achievement $o_t$2 (Park et al., 19 Aug 2025).

The evaluation setup uses 100 episodes, reported as means and standard deviations over five 20-episode chunks. Agents receive pixel observations, while the expert policy used for comparison is evaluated with state-based observations (Park et al., 19 Aug 2025). This setup is important when interpreting comparisons between imitation-learned pixel policies and the generator expert from which demonstrations were derived.

The main quantitative results are as follows.

Method	Score (%)	Return (%)
Human $o_t$3	$o_t$4	$o_t$5
Expert Policy (CrafterPlay)	97.5	$o_t$6
$o_t$7-IRIS $o_t$8	9.3	$o_t$9
prior method $x_t$0	$x_t$1	$x_t$2
C-VPT (tiny)	$x_t$3	$x_t$4
C-VPT (base)	$x_t$5	$x_t$6
C-VPT (large)	$x_t$7	$x_t$8

These results establish several points. First, CrafterVPT beats all prior baselines on both Crafter Score and Return. Second, the best Crafter Score is achieved by C-VPT (large) at 61.4, while the best Return is achieved by C-VPT (base) at 71.8 (Park et al., 19 Aug 2025). Third, the paper highlights an improvement of up to 29.6% in Crafter Score over the strongest prior method cited (Park et al., 19 Aug 2025).

This evaluation supports the central claim that CrafterVPT functions as a strong behavioral foundation for Crafter. At the same time, the gap to the Expert Policy (CrafterPlay) remains large, which is consistent with imitation from pixel observations being a compressed behavioral prior rather than a direct reproduction of the generator expert.

6. Emergent behaviors, downstream use, and limitations

The paper emphasizes that CrafterVPT learns more than achievement-specific traces. Reported emergent behaviors include constructing shelters for protection, placing items to block hostile mobs, building bridges or paths over water/lava, and strategic waiting and sleeping for safety (Park et al., 19 Aug 2025). These behaviors are presented as evidence that the model has acquired reusable survival and construction priors rather than merely memorizing task-specific action sequences.

Within downstream systems, CrafterVPT is used in two main ways. For long-horizon tasks, the pretrained C-VPT (base) is fine-tuned with PPO + LoRA using LoRA settings $x_t$9, $\tilde{x}_{1:t}$0, and a KL regularization weight of 0.1, yielding C-VPT-RL (Park et al., 19 Aug 2025). Separately, CrafterSteve-1 conditions the CrafterVPT backbone on goal embeddings from CrafterCLIP, so that the pretrained low-level policy becomes an instruction-following controller (Park et al., 19 Aug 2025). In both cases, CrafterVPT functions as the reusable behavioral backbone.

The limitations discussed in the paper are correspondingly specific. First, no-op imbalance can harm learning, but overly aggressive filtering can remove useful waiting behavior (Park et al., 19 Aug 2025). Second, behavior cloning only captures observed expert behavior; the model is strong as a prior, but not sufficient for planning-heavy tasks by itself (Park et al., 19 Aug 2025). Third, the unconditional policy struggles in hierarchical experiments on harder sequential tasks, especially T3: make wood pickaxe then obtain sapling and T4: obtain coal, craft wood pickaxe, then obtain stone (Park et al., 19 Aug 2025). Fourth, the paper explicitly shows that CrafterVPT helps but needs instruction-following and planning layers on top for multi-step goals (Park et al., 19 Aug 2025).

A common misconception would be to treat CrafterVPT as a complete general-purpose Crafter agent. The evidence presented does not support that interpretation. The model is instead a high-quality low-level prior whose principal significance lies in enabling broader embodied-agent pipelines in a lightweight environment (Park et al., 19 Aug 2025). In that sense, its contribution is architectural as well as empirical: it establishes the behavioral substrate on which the rest of CrafterDojo is built.