REVIEW 24 cited by
Consistency Policy: Accelerated Visuomotor Policies via Consistency Distillation
Not yet reviewed by Pith; the record is open.
This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.
SPECIMEN: schema-true, not a live event
T0 review · schema-true
One-sentence machine reading of the paper's core claim.
pith:XXXXXXXX · record.json · timestamp
Signed reviews
read the original abstract
Many robotic systems, such as mobile manipulators or quadrotors, cannot be equipped with high-end GPUs due to space, weight, and power constraints. These constraints prevent these systems from leveraging recent developments in visuomotor policy architectures that require high-end GPUs to achieve fast policy inference. In this paper, we propose Consistency Policy, a faster and similarly powerful alternative to Diffusion Policy for learning visuomotor robot control. By virtue of its fast inference speed, Consistency Policy can enable low latency decision making in resource-constrained robotic setups. A Consistency Policy is distilled from a pretrained Diffusion Policy by enforcing self-consistency along the Diffusion Policy's learned trajectories. We compare Consistency Policy with Diffusion Policy and other related speed-up methods across 6 simulation tasks as well as three real-world tasks where we demonstrate inference on a laptop GPU. For all these tasks, Consistency Policy speeds up inference by an order of magnitude compared to the fastest alternative method and maintains competitive success rates. We also show that the Conistency Policy training procedure is robust to the pretrained Diffusion Policy's quality, a useful result that helps practioners avoid extensive testing of the pretrained model. Key design decisions that enabled this performance are the choice of consistency objective, reduced initial sample variance, and the choice of preset chaining steps.
Forward citations
Cited by 24 Pith papers
-
Spatial Attention: Adapting Execution Horizons for Diffusion Policies via Observation Sensitivity
Under a fixed sampling budget, execution horizons that minimize disturbance-induced likelihood drop should shorten as Spatial Attention rises; forecasting it yields higher success rates than fixed horizons.
-
FA-RDP: A Frequency-Adaptive Reactive Diffusion Policy for Contact-Rich Manipulation
A shared visual-force diffusion policy with a multimodality indicator and manifold consistency distillation raises contact-rich task success to 81.7% while keeping diverse pre-contact modes.
-
SegDiff: Segmented Trajectory Diffusion for Consistent and Adaptive Robot Manipulation
SegDiff predicts continuous trajectories anchored to the next keypose and uses DDIM inversion for dynamic temporal ensembling, outperforming continuous and keypose baselines on RLBench, RoboMimic, and five real tasks.
-
High-Fidelity One-Step Generative Visuomotor Policy via Recursive Correction, Frequency Consistency, and Contrastive Flow Matching
One-step flow-matching visuomotor policy with recursive correction, dual-timestep spectral consistency, and contrastive mode separation matches or exceeds 10-step baselines at 1 NFE.
-
SeedPolicy: Horizon Scaling via Self-Evolving Diffusion Policy for Robot Manipulation
SeedPolicy introduces self-evolving gated attention to extend the temporal horizon of diffusion policies, yielding 36.8% and 169% relative gains over standard DP on clean and randomized RoboTwin 2.0 tasks.
-
TIDAL: Temporally Interleaved Diffusion and Action Loop for High-Frequency VLA Control
TIDAL raises VLA control feedback from ~2.4 Hz to ~9 Hz by caching semantic intent and interleaving one-step flow generation with execution, doubling dynamic interception success in simulation.
-
DASIP: Dynamic Test-Time Compute Scaling for Robot Control with Stochastic Interpolant Policies
A difficulty classifier adaptively selects step count, solver, and ODE/SDE mode for stochastic-interpolant robot policies, reporting 2.6–4.4x compute savings with roughly unchanged success rates.
-
RwoR: Generating Robot Demonstrations from Human Hand Collection for Policy Learning without Robot
A generative model and wrist camera turn human hand videos into robot gripper demonstrations that train manipulation policies at success rates close to those trained on real gripper data.
-
DemoSpeedup: Accelerating Visuomotor Policies via Entropy-Guided Demonstration Acceleration
DemoSpeedup accelerates visuomotor policies by downsampling high-entropy segments of demonstrations, achieving roughly 2x faster execution with maintained or improved success rates.
-
ChatVLA-2: Vision-Language-Action Model with Open-World Embodied Reasoning from Pretrained Knowledge
ChatVLA-2 uses dynamic mixture-of-experts and a two-stage training recipe to let a vision-language-action model retain pretrained reasoning while following robot instructions.
-
H$^3$DP: Triply-Hierarchical Diffusion Policy for Visuomotor Learning
H3DP couples depth-layered, multi-scale visual features to coarse-to-fine denoising stages, reporting a +27.5% relative success-rate improvement over DP3 across 44 simulation tasks.
-
TeLoGraF: Temporal Logic Planning via Graph-encoded Flow Matching
A GNN-encoded flow matching model learns to generate STL-satisfying trajectories across five robot simulation domains, with a 200K-specification dataset, reporting best-of-1024 satisfaction rates.
-
MMInference: Accelerating Pre-filling for Long-Context VLMs via Modality-Aware Permutation Sparse Attention
MMInference speeds up long-context VLM prefill by up to 8.3x at 1M tokens using modality-aware permutation sparse attention while keeping accuracy close to full attention.
-
Dynamic Rank Adjustment in Diffusion Policies for Efficient and Flexible Training
By dynamically freezing low-rank singular components of diffusion policy weights during training, DRIFT-DAgger cuts training time by roughly 11 to 18 percent while keeping task success near full-rank baselines.
-
CARP: Visuomotor Policy Learning via Coarse-to-Fine Autoregressive Prediction
A coarse-to-fine autoregressive policy with multi-scale action tokenization matches or beats diffusion policies on robot manipulation benchmarks at roughly 10x lower inference cost.
-
Diffusion-VLA: Generalizable and Interpretable Robot Foundation Model via Self-Generated Reasoning
A robot policy generates its own language reasoning before acting and injects it into a diffusion action decoder, outperforming several VLA baselines on real-robot manipulation.
-
SkillMemo: Expert-guided Skill Memory Framework for Compositional Embodied Manipulation
SkillMemo couples MoE-based skill discovery with episodic memory retrieval and reports consistent success-rate gains on diffusion and VLA policies for simulated and real manipulation tasks.
-
SeFA-Policy: Fast and Accurate Visuomotor Policy Learning with Selective Flow Alignment
Selective Flow Alignment replaces reflow-generated actions with nearby expert actions during training, yielding a one-step flow policy that beats diffusion baselines on 66 simulated and 7 real tasks.
-
Predictive Planner for Autonomous Driving with Consistency Models
A consistency-model-based predictive planner generates joint ego and agent trajectories in four sampling steps, with an alternating guided-sampling scheme to satisfy planning constraints.
-
FlowPolicy: Enabling Fast and Robust 3D Flow-based Policy via Consistency Flow Matching for Robot Manipulation
A consistency flow matching policy conditioned on 3D point clouds generates robot actions in a single inference step, running 7x faster than DP3 with comparable success rates.
-
Accelerating Human-Aware Robot Trajectory Generation via Diffusion and Consistency Distillation
A consistency-distilled diffusion model generates 150 collision-aware joint trajectories for a 6-DoF HRI manipulator in about 95 ms, with a jerk penalty improving smoothness.
-
3D-CovDiffusion: 3D-Aware Diffusion Policy for Coverage Path Planning
A diffusion policy generates ordered spray-painting trajectories from point clouds, but its claimed coverage advantage reverses against the paper's own strongest baseline on three of four categories.
-
Detecting Reading-Induced Confusion Using EEG and Eye Tracking
Multimodal EEG plus eye tracking classifies reading-induced confusion at 77.3% average weighted accuracy, beating unimodal models by 4-22%, in an 11-participant study.
-
Retrieve-Augmented Generation for Speeding up Diffusion Policy without Additional Training
RAGDP accelerates pretrained diffusion policies by initializing denoising from the nearest retrieved expert demonstration action, improving accuracy-versus-speed trade-offs without extra training.
Discussion (0). Continue with ORCID to comment.