We study this question in the representative vision-language-action (VLA) design paradigm, as implemented in architectures such as ORION, whose structure makes this question sharp under a trajectory-space logit lens. The model compresses its entire plan into a single hidden state, called the planning token, that a generative planner decodes into future waypoints. Because this vector is the sole interface between language reasoning and action, we can ask exactly when, along the decoder, the information the planner needs becomes available.
Our method reads the planning token out of every intermediate layer and decodes it with the model’s native planner, scoring the resulting trajectory. This is a trajectory-space analogue of the logit lens. Rather than measuring what a layer could encode with a trained readout, we ask what the model’s own downstream module makes of each layer. Alongside the trajectory, we read the discrete navigation intent with a linear probe.
Our empirical findings indicate that semantic intent and native-planner compatibility follow different depth-wise schedules. The navigation command is linearly decodable from the first decoder block and remains highly decodable throughout the network, despite the command never being provided to the model as textual input. By contrast, the same intermediate planning token becomes progressively more compatible with the frozen native planner, reaching its lowest trajectory error only at the final hidden state. Learned layer-1 readouts recover much of the gap, showing that the frozen-planner curve should be interpreted as a measure of representation compatibility rather than the first presence of geometric information. We extend this analysis across the five Bench2Drive ability categories, where the progression of planner compatibility is consistent.
The depth-wise diagnostic also motivates a pruning test: layers that minimally rotate the planning token may be removable even when their isolated native-planner readout is poor. We rank layers by the angular deviation they induce in the planning token, prune the decoder to trace an accuracy-latency frontier paired with the measured wall-clock of a physically shortened decoder, and break the error down by ability category. At the evaluated sample size, no family-specific degradation is statistically resolved.
Contributions
- A trajectory-space native-head lens for measuring how intermediate planning-token representations become compatible with the deployed frozen planner, alongside a linear probe of navigation intent.
- A depth-wise analysis across five Bench2Drive ability categories, showing early linear command decodability and a consistent progression of planner compatibility across the evaluated ORION checkpoint.
- A pruning study showing that planning-token input-output cosine is a better removal criterion than isolated intermediate decode quality: 8 of 32 layers can be removed within an approximately 5% relative open-loop error increase, yielding a measured 1.33-times decoder speedup.
Where Intent and Planner Compatibility Develop
The planning token carries two kinds of information that become available at different depths. The command probe is at chance at hidden state 0, where the appended planning token is still context-free. Accuracy rises to 97.7% after the first decoder block and peaks at 98.6%. Because the command is not provided in the prompt, this shows that command-related structure becomes linearly accessible after interaction with the multimodal context.
Frequently Asked Questions
What is the planning token? The planning token is a single hidden state into which the model compresses its entire plan before a generative planner decodes future waypoints.
How is intermediate planner compatibility measured? The planning token is read from every intermediate layer and decoded with the model’s native planner, with the resulting trajectory then scored.
When does navigation intent become linearly decodable? It becomes linearly decodable after the first decoder block, when accuracy rises to 97.7%, and later peaks at 98.6%.
How many decoder layers can be removed in the pruning study? Eight of 32 layers can be removed within an approximately 5% relative open-loop error increase, producing a measured 1.33-times decoder speedup.