Skip to content
RobotWorld
← Back to Videos

Proxy Policy Steering: Adapting a Frozen Generalist Policy at Inference

Loading video
Loading video

Proxy Policy Steering trains a reference proxy and a task proxy whose calibrated velocity-space residual steers a frozen base sampler at every denoising step. Across 8 real-world and 4 simulation tasks it lifts the pi-0.5 base by 53% absolute success on average while keeping broad capabilities such as failure recovery.

Robot foundation modelsPolicy AdaptationInference-Time SteeringRobot ManipulationDiffusion policiespi-0.5Manipulation
Category: research
Author: @TritiumAc
Date: 2026-09-11T00:00:00
Duration: 13.568s
Source: https://x.com/TritiumAc/status/2098442576688726518
Reference: https://arxiv.org/abs/2609.09148

Related Videos

Skild AI S1 Learns Unseen Tasks From a Single Video Demonstration
0:17

Skild AI S1 Learns Unseen Tasks From a Single Video Demonstration

@NVIDIARobotics · 615 views · 2026-09-10
Robot foundation modelsSkild AI
Dressing in Motion: Robot-Assisted Dressing While the Arm Keeps Moving
1:54

Dressing in Motion: Robot-Assisted Dressing While the Arm Keeps Moving

@robotsdigest · 583 views · 2026-09-07
Robot-Assisted DressingManipulation
Flex-π: compute-flexible robot foundation model
0:10

Flex-π: compute-flexible robot foundation model

@yokoshaan · 523 views · 2026-08-14
VLAFlex-π
Awesome-UMI Adds 9 New Device Projects
0:28

Awesome-UMI Adds 9 New Device Projects

@ArataHorie · 959 views · 2026-09-26
RoboticsImitation Learning
GPT-6 Astra Tops Independent Robot Manipulation Eval at 70.5%
0:07

GPT-6 Astra Tops Independent Robot Manipulation Eval at 70.5%

@yokoshaan · 632 views · 2026-09-21
VLARobot Manipulation
Real-Time EXPO-FT: Fast, Reliable RL for Real-Time VLA Policies
1:00

Real-Time EXPO-FT: Fast, Reliable RL for Real-Time VLA Policies

@perryadong · 561 views · 2026-09-21
VLAReinforcement Learning

As an Amazon Associate, we earn from qualifying purchases.