Skip to content
RobotWorld
← Back to Videos

RD-VLA Scales Robot Reasoning in Latent Space

Loading video
Loading video

RD-VLA uses a weight-tied recurrent action head to refine robot actions in latent space, avoiding explicit reasoning tokens while allocating test-time compute adaptively through convergence.

VLALatent ReasoningRecurrent NetworksTest-Time ComputeRobot ManipulationRD-VLAVisual-Language-Action
Category: research
Author: @DJiafei
Date: 2026-02-10T00:00:00
Duration: 82.9s
Source: https://x.com/DJiafei/status/2021250248082092094
Reference: https://rd-vla.github.io/

Related Videos

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 · 633 views · 2026-09-21
VLARobot Manipulation
Awesome-UMI Adds 9 New Device Projects
0:28

Awesome-UMI Adds 9 New Device Projects

@ArataHorie · 959 views · 2026-09-26
RoboticsImitation Learning
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
Opus 5 fails a simulated toxic-lab safety task
0:39

Opus 5 fails a simulated toxic-lab safety task

@ariansadeg · 742 views · 2026-09-19
Physical AI SafetyVLA
Show-Harness: Letting a VLM Agent Play Real Robots
1:50

Show-Harness: Letting a VLM Agent Play Real Robots

@ZechenBai · 958 views · 2026-09-10
VLAVLM Agent
VLAct: Representation-Centric Continued Pre-training for VLA Models
0:28

VLAct: Representation-Centric Continued Pre-training for VLA Models

@HuggingPapers · 783 views · 2026-08-31
VLActVLA

As an Amazon Associate, we earn from qualifying purchases.