{"id":"fde3caf0-d095-4a46-996d-54e8a28313a3","arxiv_id":"2604.19102","paper_version":1,"verdict":"UNVERDICTED","confidence":"LOW","novelty_score":6.0,"correctness_risk":"unknown","formal_verification":"none","parameter_count":2,"one_line_summary":"Selective AMP in RL enables a single policy for five humanoid gaits with faster convergence and better performance on stability tasks without losing dynamic agility.","lead":"The paper introduces a selective Adversarial Motion Prior strategy within a unified reinforcement learning framework to train a humanoid robot on five distinct gaits: walking, goose-stepping, running, stair climbing, and jumping. Smart generalists might read it to see practical advances in making robots handle both stable and dynamic movements with one policy.","discovery_kind":"new_application","skeptic_critique":{"model":"grok-4.3","headline":"The selective strategy's effectiveness depends on an ad-hoc gait classification that groups running with jumping for AMP omission, despite running being periodic like the stability gaits.","rationale":"This directly targets the reader's weakest assumption about the gait classification being accurate and sufficient. The quantitative results still demonstrate the reported improvements for the chosen selective approach, but the claim that selective is the right way because uniform over-constrains dynamic gaits would be weakened if AMP can be applied to running without issue. This is an internal assumption in the argument rather than external consensus. The test would settle if the selective choice is uniquely beneficial or if a different selection strategy could work.","tokens_in":1727,"tokens_out":328,"duration_ms":65289,"concrete_test":"Train an additional policy variant with AMP applied to running (using the same reference motion data as other periodic gaits or appropriate dynamic references), and measure convergence speed, tracking error, and success rate for the running gait; if performance is equal or superior to the selective version, the omission is not necessary to preserve agility.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The paper's motivation for selective AMP is that it would over-constrain highly dynamic gaits, but running is listed as dynamic while being periodic and stability-involved in practice. No analysis is provided on why AMP is omitted for running specifically or what would happen if AMP was applied using running reference motions. This makes the central claim's interpretation of 'without sacrificing the agility' dependent on this partitioning being optimal, which is not verified beyond the reported comparisons to uniform AMP.","agreement_with_reader":"agree"},"referee_report":{"model":"grok-4.3","summary":"The manuscript presents a reinforcement learning method for training a unified policy on a 12-DOF humanoid robot to perform five gaits: walking, goose-stepping, running, stair climbing, and jumping. The core innovation is a selective application of Adversarial Motion Prior (AMP), applied only to periodic stability-critical gaits (walking, goose-stepping, stair climbing) to accelerate convergence and reduce erratic behavior, while omitted for dynamic gaits (running, jumping) to avoid over-constraining agility. Training uses PPO with domain randomization in simulation, followed by zero-shot sim-to-real transfer. The authors claim that this selective AMP outperforms a uniform AMP baseline across all gaits in terms of convergence speed, tracking error, and success rates.","tokens_in":1820,"tokens_out":427,"duration_ms":44043,"significance":"If the empirical results hold with proper ablations and the selective strategy is robustly justified, this work could contribute to scalable multi-skill locomotion by balancing motion prior regularization with the need for dynamic expressiveness in RL policies for humanoids. The zero-shot hardware deployment adds practical value.","major_comments":[{"comment":"The central claim that selective AMP preserves agility for dynamic gaits without sacrificing performance depends on the gait classification that treats running as dynamic (AMP omitted) despite its periodic nature, similar to the stability-critical gaits. No analysis, ablation, or discussion addresses the effect of applying AMP to running reference motions, alternative partitioning, or the sensitivity of results to the selection criteria. This is load-bearing for interpreting the outperformance and the 'without sacrificing agility' assertion.","section":"Method section describing selective AMP and gait classification"}],"minor_comments":[{"comment":"The abstract asserts quantitative outperformance (faster convergence, lower tracking error, higher success rates) but supplies no specific metrics, error bars, or references to tables/figures. These details should be added or explicitly linked for immediate evaluation.","section":"Abstract"}],"recommendation":"major_revision","confidential_remarks":null},"author_rebuttal":{"model":"grok-4.3","summary":"We thank the referee for the constructive feedback on our selective AMP strategy. The major comment raises a valid point about the gait classification and lack of explicit analysis for running. We address this directly below, agreeing to strengthen the manuscript with additional discussion while defending the current approach on the basis of the existing uniform AMP baseline.","responses":[{"response":"We acknowledge that running exhibits periodicity. However, our classification prioritizes empirical training dynamics: AMP's regularization on running reference motions constrains the policy's capacity to exceed reference velocities and adapt foot placements under high-speed conditions, leading to reduced agility. The uniform AMP baseline already functions as the relevant ablation, as it applies the prior to running (and all other gaits) and yields measurably worse convergence, tracking error, and success rates on dynamic tasks compared with the selective variant. We will add a new paragraph in the Method section explicitly justifying the stability-critical versus dynamic partitioning with reference to observed policy behavior during training. We will also note the absence of exhaustive alternative partitioning experiments and sensitivity sweeps as a limitation, while arguing that the current criteria are robustly supported by the performance gap versus the uniform baseline.","revision_made":"partial","referee_comment":"The central claim that selective AMP preserves agility for dynamic gaits without sacrificing performance depends on the gait classification that treats running as dynamic (AMP omitted) despite its periodic nature, similar to the stability-critical gaits. No analysis, ablation, or discussion addresses the effect of applying AMP to running reference motions, alternative partitioning, or the sensitivity of results to the selection criteria. This is load-bearing for interpreting the outperformance and the 'without sacrificing agility' assertion."}],"tokens_in":1336,"tokens_out":358,"duration_ms":37478,"standing_objections":[]},"desk_editor":{"model":"grok-4.3","letter":"The main point is that they train one policy to handle walking, goose-stepping, running, stair climbing, and jumping on a humanoid, using the same network, actions, and rewards for all. AMP gets applied only to the first three to speed convergence and reduce bad motions, while running and jumping skip it to keep agility intact. They train in simulation with PPO and domain randomization, then transfer zero-shot to a physical 12-DOF robot. That unified setup and the hardware result are the practical parts that work.","headline":"Selective AMP on the stable gaits speeds up training in a single policy for five humanoid motions, but the decision to drop it on running rests on an untested split.","tokens_in":2312,"tokens_out":184,"would_cite":false,"duration_ms":32752,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"grok-4.3","headline":"Selective application of an adversarial motion prior in one reinforcement learning policy lets a humanoid master five gaits with faster convergence and no loss of agility.","keywords":["multi-gait learning","reinforcement learning","adversarial motion prior","humanoid locomotion","selective regularization","PPO","sim-to-real transfer","robot gaits"],"falsifier":"A uniform AMP policy trained under identical conditions that reaches equal or lower tracking error and equal or higher success rates on running and jumping, or a selective policy that produces visibly erratic or unstable running and jumping motions.","tokens_in":2609,"feed_emoji":"🤖","tokens_out":726,"duration_ms":33807,"temperature":0.7,"pith_summary":"The paper sets out to prove that a single policy structure, action space, and reward formulation can produce five distinct humanoid gaits when an adversarial motion prior is applied only to the periodic, stability-critical ones. The authors show that deliberately omitting the prior from running and jumping prevents over-constraint while still accelerating learning and reducing error on walking, goose-stepping, and stair climbing. This matters because it offers a concrete way to reconcile the stability-agility trade-off inside one training run rather than training separate controllers. Policies are trained with PPO plus domain randomization in simulation and transferred zero-shot to a physical 12-DOF robot.","feed_headline":"Selective prior lets one policy master five robot gaits","feed_subtitle":"Omitting motion regularization on running and jumping keeps agility while speeding convergence and cutting errors on stable gaits, enabling ","key_machinery":"The selective Adversarial Motion Prior strategy, which adds discriminator-based style regularization only to periodic stability-critical gaits and omits it from dynamic ones to preserve expressiveness.","core_discovery":"By applying the adversarial motion prior discriminator only to walking, goose-stepping, and stair climbing while leaving it out for running and jumping, the selective strategy produces faster convergence, lower tracking error, and higher success rates than a uniform AMP baseline across all five gaits, all within an otherwise identical PPO training setup that transfers directly to hardware.","pith_inferences":["The same selective-regularization logic could be tested on other multi-behavior robotic tasks where some skills benefit from motion-style guidance and others need unrestricted exploration.","An automatic classifier that decides on the fly whether to apply the prior might remove the need for hand-labeled gait categories.","Combining selective AMP with terrain-aware rewards could extend the approach to outdoor or uneven surfaces without separate policies.","Scaling the method to humanoids with higher degrees of freedom would test whether the selective benefit persists when action spaces grow larger."],"forward_implications":["Training time decreases and final performance improves on walking, goose-stepping, and stair climbing compared with uniform AMP.","Running and jumping retain the same agility and success rates as policies trained without any AMP term.","The identical policy architecture and reward terms suffice for all five gaits without per-gait retuning.","Zero-shot sim-to-real transfer succeeds for the full set of gaits on a 12-DOF humanoid.","Stability-focused gaits show suppressed erratic behavior while dynamic gaits keep required explosiveness."],"fun_headline_variants":["Selective prior unifies five robot gait training","Single policy learns five gaits with selective AMP","Targeted AMP enables multi-gait humanoid training","Selective motion prior for five robot gaits","One policy achieves five gaits using selective prior"],"cache_read_input_tokens":64,"weakest_assumption_plain":"That manually classifying gaits into stability-critical versus dynamic groups is accurate enough that omitting the prior from the dynamic group will preserve agility without introducing new instabilities.","fun_headline_variants_meta":{"raw":{"variants":["Selective prior unifies five robot gait training","Single policy learns five gaits with selective AMP","Targeted AMP enables multi-gait humanoid training","Selective motion prior for five robot gaits","One policy achieves five gaits using selective prior"]},"model":"grok-4.3","cost_usd":0.011454,"raw_usage":{"total_tokens":4927,"prompt_tokens":637,"num_sources_used":0,"completion_tokens":67,"cost_in_usd_ticks":114540500,"prompt_tokens_details":{"text_tokens":637,"audio_tokens":0,"image_tokens":0,"cached_tokens":64},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":4223,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":637,"tokens_out":67,"duration_ms":49563,"temperature":1.0,"reasoning_tokens":4223,"cache_read_input_tokens":64,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-05-10T02:53:08.361635+00:00","model_set":{"reader":"grok-4.3"},"falsifier":"A uniform AMP policy trained under identical conditions that reaches equal or lower tracking error and equal or higher success rates on running and jumping, or a selective policy that produces visibly erratic or unstable running and jumping motions.","supporting_citations":[],"review_version":1}