Proxy Policy Steering

• URL: http://arxiv.org/abs/2609.09148v3
• Date: Sun, 13 Sep 2026 20:53:27 GMT
• Title: Proxy Policy Steering
• Abstract summary: Generalist robot policies carry broad manipulation priors from large-scale data, but specializing them to a new task remains the deployment bottleneck.<n>We introduce Proxy Policy Steering (PPS), an inference-time adaptation method that resolves this challenge.<n>PPS lifts the state-of-the-art pi 0.5 base policy by 53% absolute success rate on average, with zero-to-one gains on tasks the base never solves.
• Score: 18.880191675273796
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Generalist robot policies carry broad manipulation priors from large-scale data, but specializing them to a new task remains the deployment bottleneck. This requires eliciting task-specific behavior from limited demonstrations without degrading their broad capabilities. We introduce Proxy Policy Steering (PPS), an inference-time adaptation method that resolves this challenge by training two lightweight proxy policies whose calibrated velocity-space difference steers the frozen base sampler. 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. Because the base is never directly modified, its broad capabilities remain available at inference, including behaviors such as recovery from failure that the demonstrations themselves do not exercise. Adaptation requires only forward velocity predictions from the base, making PPS lightweight to train and applicable even without access to the base's parameters. On 8 real-world and 4 simulation manipulation tasks, PPS lifts the state-of-the-art pi 0.5 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.

Related papers

• ORPA: Online Residual Policy Adaptation for Robot Manipulation Control with Human Feedback [1.711436622393758]
  Online Residual Policy Adaptation (ORPA) is a framework that enables feedback-driven correction of robot actions without modifying the underlying policy parameters.<n>ORPA augments a pretrained control policy with a lightweight, feedback-conditioned module that predicts residual adjustments directly in joint space.<n>We evaluate ORPA on a set of precision-sensitive manipulation tasks using the ALOHA platform, demonstrating improvements in success rate and recovery from small perturbations compared to baseline control policies and rule-based inverse kinematics corrections.
  arXiv  Detail & Related papers  (2026-08-18T03:30:12Z)
• Behavior Uncloning: Distilling Mode Redirection into Policy Weights without Inference-Time Steering [28.889110668832984]
  We propose MoRE, which redirects policy rollouts toward desired behavior modes through a short "uncloning" step.<n>MoRE improves the average deployment success rate (SR) by 44 percentage points over the original mixed-mode policy.
  arXiv  Detail & Related papers  (2026-06-28T05:01:27Z)
• Test-Time Gradient Guidance of Flow Policies in Reinforcement Learning [50.738952715864116]
  Expressive continuous control policies form the backbone of advances in scaling imitation learning for simulated and real robot control.<n>We propose QGF (Q-Guided Flow), an RL algorithm that performs policy optimization entirely at test time.<n> Empirically, QGF outperforms prior test-time RL methods on single-task and goal-conditioned offline RL benchmarks.
  arXiv  Detail & Related papers  (2026-06-09T16:45:57Z)
• One-Way Policy Optimization for Self-Evolving LLMs [63.8638342097375]
  Reinforcement Learning with Verifiable Rewards (RLVR) has become a promising paradigm for scaling reasoning capabilities of Large Language Models (LLMs)<n>We propose One-Way Policy Optimization (OWPO), a method based on the principle of decoupling optimization direction from update magnitude.<n> Experimental results demonstrate that OWPO outperforms strong baselines, including DAPO, OPD, and MOPD.
  arXiv  Detail & Related papers  (2026-05-21T08:25:27Z)
• Beyond Action Residuals: Real-World Robot Policy Steering via Bottleneck Latent Reinforcement Learning [42.74142065376427]
  We propose Z-Perturbation Reinforcement Learning (ZPRL) to steer pretrained policies through a compact bottleneck latent.<n>In the real world, ZPRL improves the average success rate on four tasks by 33.7% over imitation base policies.
  arXiv  Detail & Related papers  (2026-05-19T14:43:26Z)
• Trust the Batch, On- or Off-Policy: Adaptive Policy Optimization for RL Post-Training [50.86545293331458]
  Reinforcement learning is structurally harder than supervised learning because the policy changes the data distribution it learns from.<n>We present a simple yet effective batch-adaptive objective that replaces fixed clipping with the normalized effective sample size of the policy ratios.
  arXiv  Detail & Related papers  (2026-05-12T16:44:47Z)
• OGPO: Sample Efficient Full-Finetuning of Generative Control Policies [53.42266064673132]
  Generative control policies (GCPs) have emerged as effective parameterizations for robot learning.<n>This work introduces Off-policy Generative Policy Optimization (OGPO), a sample-efficient for finetuning GCPs.<n>OGPO achieves state-of-the-art performance on manipulation tasks spanning multi-task settings, high-precision insertion, and dexterous control.
  arXiv  Detail & Related papers  (2026-05-04T18:36:40Z)
• Speedup Patch: Learning a Plug-and-Play Policy to Accelerate Embodied Manipulation [19.765462206627955]
  Speedup Patch (SuP) is a policy-agnostic framework that enables plug-and-play acceleration using solely offline data.<n>SuP achieves an overall 1.8x execution speedup for diverse policies while maintaining their original success rates.
  arXiv  Detail & Related papers  (2026-03-21T05:29:04Z)
• When to Act, Ask, or Learn: Uncertainty-Aware Policy Steering [10.01278648231868]
  Policy steering is an emerging way to adapt robot behaviors at deployment-time.<n> Vision-Language Models (VLMs) are promising general-purpose verifiers due to their reasoning capabilities.<n>We propose uncertainty-aware policy steering (UPS), a framework that jointly reasons about semantic task uncertainty and low-level action feasibility.
  arXiv  Detail & Related papers  (2026-02-25T23:23:22Z)
• Off-Policy Evaluation for Large Action Spaces via Policy Convolution [60.6953713877886]
  Policy Convolution family of estimators uses latent structure within actions to strategically convolve the logging and target policies. Experiments on synthetic and benchmark datasets demonstrate remarkable mean squared error (MSE) improvements when using PC.
  arXiv  Detail & Related papers  (2023-10-24T01:00:01Z)
• Offline Reinforcement Learning with On-Policy Q-Function Regularization [57.09073809901382]
  We deal with the (potentially catastrophic) extrapolation error induced by the distribution shift between the history dataset and the desired policy. We propose two algorithms taking advantage of the estimated Q-function through regularizations, and demonstrate they exhibit strong performance on the D4RL benchmarks.
  arXiv  Detail & Related papers  (2023-07-25T21:38:08Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.