VLA

VLA directly connects pre-trained vision-language models with robot action policies, enabling robots to process natural language instructions, camera images, and low-level motion control in a unified manner.

Compared with traditional end-to-end imitation learning, VLA models have stronger generalization capabilities, but high inference latency, stringent requirements for training data diversity, and deployment on real-time control and low-cost hardware remain the primary challenges.

VLA models integrate visual perception, language understanding, and robot action planning into a unified model, typically using a pretrained large vision-language model as the backbone, with representative works including Google's RT-2 and OpenVLA.

The model belongs to the category of Vision-Language-Action (VLA) models, unifying a visual encoder, a language understanding module, and an action prediction head within an end-to-end architecture.

The current mainstream technical approach to embodied intelligence is the Vision-Language-Action model (VLA), which takes camera images and language instructions as joint inputs and directly outputs joint control signals for robotic arms.

RT-2 first demonstrated the feasibility of the VLA architecture by discretizing robot actions into tokens and having the vision-language model directly output action sequences.

DeepMind developed the VLA (Vision-Language-Action model) system, which treats the robot's physical actions on equal footing with visual tokens and language tokens.

Google DeepMind的RT-2首次将视觉-语言模型直接用于生成机器人动作指令

AtomicRL is a reinforcement learning post-training framework that leverages atomic task structure to achieve reliable credit assignment and adaptive exploration, used to improve the performance of vision-language-action (VLA) models in long-horizon robotic manipulation.

On the Find-the-Item task, current state-of-the-art VLA models are only at or below chance level (success rate ≤33.3%).

Jia Xiaosong observed that models such as VLA and WAM that introduce stronger vision-language capabilities are not necessarily better than well-optimized end-to-end models on some autonomous driving benchmarks; what truly limits the model is whether it can make the right decision, not whether it recognizes a certain object.

In a closed-loop trajectory, a single action deviation puts the robot into a state outside the training distribution, and each subsequent step compounds error on top of the mistake, ultimately causing the entire episode to collapse—a failure that may show up as only a few percentage points of decline in average success rate.

CARE is a certification framework that provides statistical guarantees for VLA model acceleration configurations, guaranteeing that at least 85.8% of reference solved episodes still succeed at a user-specified failure budget with 95% confidence.

Existing VLA accelerators (such as action chunking, visual token pruning, and quantized compilation) generally report only average task success rates, masking the risk of cascading failures that can occur during closed-loop execution.

Agents built on frontier models such as Opus and Astra are several generations ahead of any VLA in high-level reasoning capability.

Once a VLA model is trained, its behavioral patterns are frozen, making it difficult to personalize and adapt to a specific user's long-term preferences, and it also lacks proactive anticipation of danger and self-recovery from failure.

The CarryOn framework adopts a dual-loop architecture, layering a planning loop and an execution loop, two loops that work in coordination, on top of a frozen VLA policy.

CarryOn's execution loop actively supervises the VLA policy's action outputs, masking actions predicted to involve dangerous contact, achieving prediction-based proactive protection.

Even when the underlying VLA has already undergone safety fine-tuning, the safety violation rate is still as high as 52.6%.

The paper's experiments show that current mainstream VLA policies perform poorly in zero-shot personalized scenarios: they can complete tasks, but it is difficult for them to follow preferences from instructions alone.