Proxy Policy Steering

Chuanruo Ning*, Tianrui Wang*, Wei-Chiu Ma, Kuan Fang Cornell University * Equal contribution     Equal advising
We propose Proxy Policy Steering (PPS), which adapts a frozen base policy πbase using two learned proxy policies, πref and πtask, combined in velocity space. By adding the proxy residual to the base prior, PPS steers the action distribution toward task behaviors while preserving general priors.
Abstract

Robotics foundation models carry broad manipulation priors from large-scale data, but specializing them to a new task remains the deployment bottleneck. This requires injecting new task-specific behavior from limited demonstrations without eroding their broad capabilities. We introduce Proxy Policy Steering (PPS), a novel adaptation method that addresses this challenge by training two lightweight proxy policies. A reference proxy models the frozen base's behavior on target-task observations, and a task proxy, initialized from the reference, captures how this behavior changes under task supervision. Their difference forms a calibrated velocity-space residual that steers the frozen base sampler at every denoising step. We identify the conditions under which this residual isolates the change induced by task supervision, and validate them empirically. On 8 real-world and 4 simulation manipulation tasks, PPS lifts the state-of-the-art π0.5 base policy by 53% absolute success rate on average, with zero-to-one gains on tasks the base never solves, while preserving the base's broad capabilities. PPS outperforms LoRA fine-tuning, from-scratch specialists, residual policies, and prior inference-time steering methods.

Method

Our key insight is that effective adaptation should neither steer only within the base's existing distribution nor replace the base with a narrowly trained specialist. Instead, PPS estimates the direction in which task supervision would steer the base's action distribution, leaving its weights and broad priors untouched.

To estimate that direction without access to base gradients, PPS trains two lightweight proxy policies and takes their difference in velocity space.

a log πbase
γa log πref
+
γa log πtask
a log πPPS
base distribution
πbase
Distill
ref distribution
πref
Finetune
task distribution
πtask
pps distribution
πPPS
rollout 1 sin
State within task distribution
Novel task behavior
rollout 2 sout
State outside task distribution
Recovery via base prior
rollout 3 sboth
State within task and base distribution
Agreement reached

At inference, PPS integrates the velocity predictions (v) of three models in each flow-matching step:

vPPS = vbase + γ · (vtask vref), vi :=a log πi

The coefficient γ sets the steering strength: when the demonstrations call for new actions the residual drives the sampler toward the task proxy, and when the proxies agree it vanishes and the frozen base stays in control, preserving broadly useful priors.

Effective Steering

We evaluate PPS on 8 real-world and 4 simulation manipulation tasks, spanning long-horizon goals, articulated and deformable objects, and diverse skills. PPS delivers consistent gains over the base model and outperforms fine-tuned models, from-scratch specialists, and prior steering baselines, with zero-to-one gains on tasks the base never solves on its own.

Three bar charts — Real-world, Simulation, and π0 transfer — comparing average success rates of π0.5 base, Specialist, Residual, DSRL, LoRA, and PPS. PPS leads every panel.
Select task to see example rollouts and comparisons
Coffee Brewing
Base: 0/10 PPS: 5/10
PPS
Success
Base (π₀.₅)
Can't open the lid
LoRA
Overfits after a near miss
PPS
Success
Base (π₀.₅)
Places instead of inserting
LoRA
Continues a wrong insertion
PPS
Success
Base (π₀.₅)
Can't fold
LoRA
Overfits, tosses the jeans
PPS
Success
Specialist
Overfits to demos
LoRA
Tips the box, can't recover
PPS
Success
Base (π₀.₅)
Releases before wiping
LoRA
Drags before the tissue is out
PPS
Success
Base (π₀.₅)
Can't open the drawer
LoRA
Stows before opening
PPS
Success
Specialist
Overfit to pressing action
LoRA
Overfit to pressing action
PPS
Success
Base (π₀.₅)
Imprecise insertion
LoRA
Overfit to insertion mode
PPS
Success
Base (π₀.₅)
Misses the placing
LoRA
Grasping failure, can't recover
PPS
Success
Base (π₀.₅)
Picks egg before opening lid
LoRA
Overfits to incorrect placing mode
PPS
Success
Base (π₀.₅)
Can't pick up the tea pot
LoRA
Missed grasp and fails to recover
PPS
Success
Base (π₀.₅)
Can't open the lid
LoRA
Missed grasp and fails to recover
Visualize the steering effect

Drag the steering strength γ and watch the trade off between base priors and task guidance.

Robust Execution

Because PPS leaves the base model untouched, the combined policy inherits its broad priors and recovers from failures it never saw in the demonstrations. LoRA fine-tuning, by contrast, overfits to the handful of demonstrations and loses these priors, breaking down under the same perturbations.

LoRA
PPS
LoRA
PPS
Analysis

Through a series of analyses, we study what makes PPS effective and where it still falls short.

Success rate versus steering scale: PPS peaks near 0.4 to 0.6, above the specialist and base baselines.

Steering strength. The coefficient γ trades base priors against task steering. Success peaks for γ in (0.4, 0.6), well above both baselines.

Proxy-training ablation: w/o ref, w/o vel, and w/o tune all trail the full method.

Proxy training. On-policy distillation, velocity-level supervision, and reference initialization each ensure the two proxies differ by task supervision alone, and their difference is an informative task signal.

Failure-mode breakdown for Specialist, base, and PPS by root cause; PPS has far fewer total failures.

Failure modes. PPS fails far less often than the specialist and base, cutting incorrect-mode errors 69% versus the base and out-of-distribution errors 44% versus the specialist.

Data scaling on Tea and Capsule: success rate versus number of demonstrations from 50 to 200 for Specialist, LoRA, and PPS.

Data scaling. On the two lowest-success tasks (Tea and Capsule), PPS leads at every demonstration budget from 50 to 200, with its largest margin in the low-data regime. The from-scratch specialist improves steadily and catches up near 200 demonstrations.

Cross-Modality Steering

Because the proxies are separate networks, they can take inputs the base never receives. PPS routes extra modalities such as point cloud or audio into the frozen, RGB-only base through the task proxy: the point cloud locates a texture-less object, and audio tells the ringing phone from the silent one. This makes PPS a lightweight way to add new sensors without retraining the base.

Additional Modality
PPS Policy Rollout
Success Rate
point cloud
Marble task success rate: PPS reaches about 82%, ahead of Specialist, LoRA, LoRA-MM, DSRL, and the base policy.
audio
Phone task success rate: PPS reaches about 86%, ahead of LoRA-MM, Specialist, LoRA, DSRL, and the base policy.
Legend: base policy, Specialist, DSRL, LoRA, LoRA-MM, PPS.
BibTeX
@article{ning2026proxy,
  title   = {Proxy Policy Steering},
  author  = {Ning, Chuanruo and Wang, Tianrui and Ma, Wei-Chiu and Fang, Kuan},
  journal = {arXiv preprint arXiv:XXXX.XXXXX},
  year    = {2026}
}