← Back to Videos

D&A Turns Any VLM Into a Robot Reward Model

Loading video

D&A separates description from assessment so any VLM family can provide stronger zero-shot robot rewards for RL, filtered imitation learning and code-as-policies feedback.

Robot Rewards VLM Zero-Shot Learning Reinforcement Learning D&A Vision-language models

Category: research

Author: @memmelma

Date: 2026-11-03T00:00:00

Duration: 33.266s

Source: https://x.com/memmelma/status/2117628112972247073

Reference: https://weirdlabuw.github.io/dna

Gym-Anything Builds Robot Training Worlds

Gym-Anything Builds Robot Training Worlds

@wellecks · 67 views · 2026-10-30

Robot simulationAgentic Coding

Draft: Parametric Robot Design Exploration Without Rebuilding CAD

Draft: Parametric Robot Design Exploration Without Rebuilding CAD

@spaceandtech_ · 341 views · 2026-10-11

Robot DesignMJCF

Exploy: Open-Source Library Bridging Sim-to-Real for RL Deployment

Exploy: Open-Source Library Bridging Sim-to-Real for RL Deployment

@rai_inst · 663 views · 2026-10-07

Sim-to-RealReinforcement Learning

RT-SAFE: real-time embodied-agent safety benchmarking

RT-SAFE: real-time embodied-agent safety benchmarking

@XTRose88 · 718 views · 2026-10-06

Embodied AISafety Evaluation

Kardashev-0.7: A Trained Swarm of 32 Models

Kardashev-0.7: A Trained Swarm of 32 Models

@MLCatttt · 489 views · 2026-10-05

Model SwarmReinforcement Learning

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

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

@perryadong · 593 views · 2026-09-21

VLAReinforcement Learning

As an Amazon Associate, we earn from qualifying purchases.