BEIJING, Oct. 9, 2026 /PRNewswire/ -- ROBOTERA's World Action Model (WAM), VPP2 (Video Prediction Policy 2), has ranked No. 1 on RoboDojo without additional data or agent-based reinforcement self-improvement (Agent RSI). Developed by the University of Hong Kong's MMLab in collaboration with nearly 20 leading academic institutions worldwide, RoboDojo is a rigorous benchmark for general-purpose robot manipulation that evaluates performance across challenging simulation and real-world tasks, going beyond simple demonstrations.

VPP2 achieved an average success rate of 32.26% and an average score of 39.26, ranking first in Generalization, Precision, and Memory among evaluated methods.

ROBOTERA has open-sourced VPP2 on GitHub, with further details available on the project website.

Building a More Generalizable World Action Model

VPP2 enables robots to better predict how actions will change their surroundings and translate instructions into physical movements. Unlike video models designed primarily to generate visual content, VPP2 is trained to understand object movements and follow precise manipulation instructions. It combines video prediction with action generation, helping robots perform tasks more reliably across different objects, environments, and scenarios.

For complex tasks involving multiple steps, VPP2 can also work with a vision-language model (VLM) that breaks down high-level instructions into smaller, executable actions.

Validated Across Simulation and Real-World Tasks

VPP2 was evaluated across video prediction, instruction following, and robotic manipulation tasks.

On the ALOHA platform, VPP2 achieved a 58.5% average success rate across 10 task categories, outperforming leading baselines in nine. It also achieved 45.0% on LIBERO-Pro, a benchmark for robotic manipulation and generalization. With high-level task planning, average success rates across five task groups more than doubled, from 27.6% to 57.6%.

These results demonstrate VPP2's ability to connect visual understanding, instruction following, and physical action, advancing general-purpose robotics toward practical applications.

From Research Breakthroughs to Real-World Deployment

VPP2 marks ROBOTERA's fourth benchmark championship in embodied intelligence in 2026, following top results at World Arena, Benjie's Humanoid Olympic Games, and RoboChallenge. Combined with ongoing humanoid-robot deployments with China Post and SF Express across more than 10 logistics centers in China, these achievements reflect ROBOTERA's progress in both advancing general-purpose robot intelligence and bringing it into real-world applications.

Building on its advances in world-action modeling and real-world deployment experience, ROBOTERA is working to make general-purpose robots reliable partners in everyday work.

View original content to download multimedia:https://www.prnewswire.co.uk/news-releases/robotera-tops-embodied-ai-benchmark-robodojo-without-additional-data-or-agent-rsi-302903611.html

Images

Tags

More Like This

Embodied Intelligence Pioneer Robotera Secures RMB 1 Billion in Series A+ Funding

Robbyant Launches LingBot-VA 2.0 Built Natively for Embodied AI and Physical World Control

ROBOTERA Raises RMB 1 Billion in Strategic Round, Valuation Exceeds RMB 10 Billion with Broad Industry Support

Robbyant Upgrades and Open-Sources LingBot-VLA 2.0 as a Next-Generation Universal Brain for Embodied AI

ROBOTERA Raises Over USD 200 Million in New Round Led by SF Group, HSG and IDG Capital

Elite Robots Collaborates with Generalist AI on Next-Gen Embodied AI

AGIBOT Unveils New Generation of Embodied AI Robots and Models, Accelerating Real-World Deployment of Physical AI

AgiBot Robotics Debuted at IROS 2025, the AgiBot World Challenge Concluded Successfully