Dexmal provides tools to accelerate vision‑language‑action (VLA) research and deployment in robotics. Its open‑source Dexbotic codebase consolidates state‑of‑the‑art VLA models, datasets, and evaluation pipelines, while the RoboChallenge cloud platform runs submitted models on real robot hardware, delivering safety‑checked performance metrics. These offerings help research labs and AI developers quickly prototype, benchmark, and validate autonomous robot systems in the physical world.
Funding
Funding not disclosed
Founders
Product
Problem
Robotics research and development often lack unified, open-source tools for training and evaluating vision‑language‑action (VLA) models, and there are limited accessible platforms for testing these models on real robotic hardware, slowing progress and increasing integration costs.
Solution
Dexmal addresses this gap by offering two complementary offerings. First, Dexbotic is an all‑in‑one open‑source codebase that consolidates state‑of‑the‑art VLA architectures, datasets, and evaluation pipelines, enabling researchers to implement, compare, and extend models within a single repository. Second, RoboChallenge provides a cloud‑based service that runs submitted models on actual robots, delivering real‑world performance metrics and safety validation. Together, these resources accelerate the development cycle from simulation to physical deployment while promoting reproducibility and collaborative advancement in robotics AI.
Target Audience
Primary users are robotics research labs, AI developers, and companies building autonomous robot systems that require robust VLA models and real‑world validation.
Features
- Dexbotic repository with modular implementations of leading VLA models and unified training/evaluation scripts
- Integrated dataset loaders and benchmark suites for vision‑language‑action tasks
- Comprehensive documentation and tech reports to facilitate rapid onboarding
- RoboChallenge online platform that executes submitted models on diverse real‑world robot platforms
- Automated safety checks and performance reporting, including latency, success rates, and failure diagnostics
- Open‑source licensing and community contribution workflow to foster collaborative improvement
- API access for programmatic submission and retrieval of test results