cd /news/robotics/demystifying-physical-ai-a-deep-dive… Β· home β€Ί topics β€Ί robotics β€Ί article
[ARTICLE Β· art-173659] src=pub.towardsai.net β†— pub= topic=robotics verified=true sentiment=Β· neutral

Demystifying Physical AI: A Deep Dive into VLA Architectures, Action Chunking, and Split Inference

Physical AI systems need specialized Vision-Language-Action (VLA) architectures because querying a 2-billion-parameter Vision-Language Model for every motor adjustment yields only a fragile 2 Hz control loop, according to a technical deep dive on VLA design. The article details how Physical Intelligence's Ο€0 and Ο€0.5 pair a roughly 400M-parameter SigLIP encoder and a Gemma 2B transformer backbone with a roughly 300M-parameter flow-matching action expert to emit action chunks at 50Hz or higher, and how split inference keeps closed-loop execution under 20ms by running multi-billion-parameter models off-board. It also contrasts flow-matching and CVAE latent style sampling (ACT) with non-generative latent-space predictive world models (VLA-JEPA).

recommendation

Decouple high-level VLM reasoning from control: cache the backbone's KV state and let a small flow-matching action head emit action chunks at 50Hz+, running the large model off-board to hold closed-loop latency under 20ms.

more in playbook: Designing agent systems
by read6 min views5 publishedNov 21, 2026

Imagine taking a state-of-the-art 2-billion-parameter Vision-Language Model (VLM) and forcing it to control a multi-joint robotic arm in real time. If you query the language backbone for every single motor adjustment, your robot stutters along at a fragile, high-latency 2 Hz β€” likely knocking over a coffee cup long before it reaches its target.

The fundamental challenge of Physical AI isn’t just multimodal understanding; it’s translating high-level semantic reasoning into high-frequency, continuous physicalΒ control.

TRADITIONAL VLM                              PHYSICAL AI (VLA)β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”                β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”β”‚ High-Level Reasoning      β”‚                β”‚ High-Level Reasoning      β”‚β”‚ Text / Image β†’ Text Tokensβ”‚                β”‚ Text / Image β†’ Embeddings β”‚β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”¬β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜                β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”¬β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜              β”‚ Low Frequency                              β”‚ Bottleneck Bridge              β–Ό                                            β–Όβ”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”                β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”β”‚ Discrete Text Output      β”‚                β”‚ Continuous Action Expert  β”‚β”‚ (Too slow for control)    β”‚                β”‚ Flow Matching / CVAE      β”‚β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜                β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”¬β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜                                                           β”‚ High Frequency (50Hz+)                                                           β–Ό                                             β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”                                             β”‚ Joint Angles / Torques    β”‚                                             β”‚ Real-Time Control Loops   β”‚                                             β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜

To bridge this gap, AI researchers and robotics engineers are moving beyond standard token generation toward specialized Vision-Language-Action (VLA) architectures. From continuous flow-matching action experts (Ο€0) and CVAE latent style sampling (ACT) to non-generative, latent-space predictive world models (VLA-JEPA), the physical AI stack is evolving at breakneck speed.

In this deep dive, we will unroll the architectural blueprints behind modern VLAs, deconstruct how multi-view vision encoders, language backbones, and continuous action heads interface, and explore how split inference enables multi-billion-parameter foundation models to run off-board while maintaining sub-20ms closed-loop execution on real hardware.

Physical Intelligence’s Ο€0 and Ο€0.5 represent a groundbreaking architectural approach: adapting pre-trained Vision-Language Models into real-time, highly dexterous controllers without losing their pre-trained semantic grounding.

β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”      β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”β”‚ Multi-View Images      β”‚      β”‚ Text Prompt &          β”‚β”‚ (RGB Cameras)          β”‚      β”‚ Proprioception         β”‚β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”¬β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜      β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”¬β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜            β”‚                               β”‚            β–Ό                               β”‚β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”                  β”‚β”‚ SigLIP Encoder (~400M) β”‚                  β”‚β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”¬β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜                  β”‚            β”‚ Image Embeddings              β”‚ Text Tokens / State            β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”¬β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜                            β–Ό           β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”           β”‚Concatenation & Linear Projectionβ”‚           β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”¬β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜                            β–Ό           β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”           β”‚ Gemma 2B Transformer Backbone  β”‚           β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”¬β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜                            β”‚ Key-Value (KV) Cache                            β–Ό           β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”           β”‚ Action Expert (~300M)          β”‚           β”‚ (Flow-Matching Decoder)        β”‚           β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”¬β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜                            β–Ό                   Action Chunk Sequences
Training Mode                     Inference / Action Mode           Action Chunking β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”        β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”     β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β” β”‚ Full Backprop Through β”‚        β”‚  KV-Cache Frozen      β”‚     β”‚  Predict N Steps Aheadβ”‚ β”‚ SigLIP + Gemma 2B +   β”‚        β”‚  Fast Re-sampling     β”‚     β”‚  [a_t, a_t+1, ... ]   β”‚ β”‚ Action Expert         β”‚        β”‚  Action Expert at 50Hzβ”‚     β”‚  smoothly overlapped  β”‚ β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜        β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜     β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜

1. Training Mode:

The system is trained end-to-end on massive, heterogeneous datasets comprising human teleoperation and multi-robot interaction logs. Trajectory paths are supervised via continuous flow-matching loss functions, aligning visual-semantic representations with physical actuation dynamics.

2. Execution / Action Mode (Queue-Based Inference):

Evaluating a full 2-billion+ parameter VLM at 50 Hz is computationally unfeasible on edge hardware. In action mode, the Gemma 2B backbone evaluates inputs at a lower frequency (e.g., 2–5 Hz), populating and updating a frozen Key-Value (KV) cache. The lightweight 300M Action Expert queries this frozen KV-cache at 50 Hz, sampling control vectors in milliseconds.

3. Action Chunking:

Instead of predicting a single action at, the Action Expert outputs an entire action chunk trajectory horizon A=[at,at+1,…,at+k] (e.g., 30 to 50 steps into the future). The robot’s low-level execution layer consumes these vector sequences from a fast localΒ queue.

Pioneered by Tony Z. Zhao et al. (Stanford ALOHA), the Action Chunking with Transformer (ACT) model directly addresses two core issues in imitation learning: compounding error during sequential execution and non-Markovian human teleoperation data.

TRAINING (CVAE Pipeline)                      β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”      β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β” β”‚ Current Observation  β”‚      β”‚ Ground Truth Action  β”‚ β”‚ (Images + Joint Pos) β”‚      β”‚ Sequence (a_{t:t+k}) β”‚ β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”¬β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜      β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”¬β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜            β”‚                             β”‚            β–Ό                             β–Όβ”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”β”‚  CVAE Encoder (Transformer / ResNet Backbone)        β”‚β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”¬β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜                           β”‚                           β–Ό  Latent Variable z ~ N(ΞΌ, Οƒ)β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”΄β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”β”‚  CVAE Decoder (Transformer Decoder)                  β”‚β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”¬β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜                           β”‚                           β–Ό              Predicted Action Chunk (a_{t:t+k})
INFERENCE MODE                     β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”       β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”β”‚ Current Observation  β”‚       β”‚  Latent Vector z     β”‚β”‚ (Images + Joint Pos) β”‚       β”‚  (Set to Mean z = 0) β”‚β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”¬β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜       β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”¬β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜           β”‚                              β”‚           β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”¬β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜                          β–Όβ”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”β”‚  CVAE Decoder (Transformer Decoder)                  β”‚β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”¬β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜                           β”‚                           β–Ό              Predicted Action Chunk (a_{t:t+k})

3. Temporal Ensemble Smoothing:

To prevent jerky movements when transitioning between action chunks, ACT re-computes overlapping predictions at every timestep t. Overlapping action predictions for the same future instant are combined using an exponential weighting scheme:

at=i=1βˆ‘K**wiβ‹…at(i)

This temporal ensembling smooths out sudden trajectory shifts, enabling precise multi-joint manipulation.

Traditional VLA models map pixels straight to actions via supervised imitation. However, pixel space contains significant task-irrelevant noise (shadows, background movement, lighting changes). The Joint-Embedding Predictive Architecture VLA (VLA-JEPA) avoids pixel generation altogether, shifting training into latent world dynamics prediction.

TRAINING ARCHITECTURE (JEPA World Model) Target Path:       β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”    β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”(Future Frame)     β”‚Future Image (x_t+k)│───>β”‚ Target Encoder  │──┐                   β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜    β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜  β”‚                                                                  β–Ό                                                   Target Latent State (s_t+k)                                                                  β”‚ Context Path:     β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”    β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”  β”‚  L_2 / Cosine Loss (Current Frame)   β”‚ Current Image (x_t)│───>β”‚ Context Encoder β”‚  β”‚  (No Pixel Decoding)                   β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜    β””β”€β”€β”€β”€β”€β”€β”€β”€β”¬β”€β”€β”€β”€β”€β”€β”€β”€β”˜  β”‚                                                      β”‚           β”‚                                                      β–Ό           β”‚                   β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”    β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”  β”‚ Proposed Action:  β”‚ Candidate Action   │───>β”‚ World Predictor │──┴─> Match Target!                   β”‚ Command (a_t)      β”‚    β”‚ Model (Transformer)                   β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜    β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜
INFERENCE ARCHITECTURE (Policy Execution)                      β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”                   β”‚ Current Image (x_t)   β”‚                   β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”¬β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜                               β”‚                               β–Ό                   β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”                   β”‚ Context Encoder       β”‚                   β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”¬β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜                               β”‚                               β–Ό Latent State (s_t)                   β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”                   β”‚ Action Head / Policy  β”‚                   β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”¬β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜                               β”‚                               β–Ό                    Action Command Sequence (a_{t:t+k})

A World Model learns the transition dynamics of physical reality: P(st+1∣st,at). Rather than predicting exact RGB values, it models how physical state representations (object affordances, spatial boundaries, movement vectors) evolve given specific motor inputs.

Training Phase (Self-Supervised Latent Dynamics):

Inference Phase (Fast Policy Execution):

Deploying massive multi-billion parameter VLA models on mobile hardware creates an engineering trade-off: onboard compute is bounded by power draw, thermal constraints, and payload weight. Yet closed-loop robot control demands low-latency execution (20ms to 1ms response loops).

The solution gaining rapid adoption across modern robotics is Split Inference.

SPLIT INFERENCE ARCHITECTURE β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β” β”‚                   OFF-BOARD SERVER / EDGE GPU CLUSTER                  β”‚ β”‚                                                                        β”‚ β”‚  β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”        β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”  β”‚ β”‚  β”‚ Camera Feeds & Goals  │───────>β”‚ Heavy VLA Backbone              β”‚  β”‚ β”‚  β”‚ (Ingested via Wi-Fi)  β”‚        β”‚ (SigLIP + Gemma-2B / VLA-JEPA)  β”‚  β”‚ β”‚  β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜        β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”¬β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜  β”‚ β”‚                                                    β”‚                   β”‚ β”‚                                                    β–Ό                   β”‚ β”‚                                  Generates 50-Step Action Chunks       β”‚ β”‚                                  at 2-5Hz Low Frequency                β”‚ β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”¬β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜                                           β”‚                        Low-Latency Network Transmission (gRPC/WebSockets)                                           β”‚ β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β–Όβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β” β”‚                      ON-BOARD EMBEDDED ROBOT COMPUTE                   β”‚ β”‚                      (e.g., NVIDIA Jetson Orin)                        β”‚ β”‚                                                                        β”‚ β”‚                                  β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”   β”‚ β”‚                                  β”‚ Action Trajectory Queue         β”‚   β”‚ β”‚                                  β”‚ (Buffers dynamic commands)      β”‚   β”‚ β”‚                                  β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”¬β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜   β”‚ β”‚                                                   β”‚                    β”‚ β”‚                                                   β–Ό                    β”‚ β”‚  β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”        β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”  β”‚ β”‚  β”‚ Onboard Micro-Policy  │───────>β”‚ High-Frequency Motor PD Control β”‚  β”‚ β”‚  β”‚ (Local Reflex Layer)  β”‚        β”‚ (100Hz–1kHz Torque Drives)      β”‚  β”‚ β”‚  β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜        β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜  β”‚ β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”΄β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜

An off-board workstation or local GPU cluster housing enterprise accelerators (e.g., NVIDIA H100/L40S) hosts the heavy Vision-Language backbone. It processes multi-view camera streams and text instructions, outputting long Action Chunks (e.g., N=50 future trajectory points) at a low rate of 2 Hz to 5Β Hz.

Computed action vectors are streamed over high-bandwidth, ultra-low-latency wireless links (Wi-Fi 6E, private 5G, or local network protocols) as lightweight structured tensor payloads.

An embedded edge device (e.g., NVIDIA Jetson Orin) maintains a synchronized Action Trajectory Queue. The robot’s internal motor control loop pops commands from this queue at 50 Hz to 1000 Hz, feeding motor drivers, torque loops, and joint controllers.

Demystifying Physical AI: A Deep Dive into VLA Architectures, Action Chunking, and Split Inference was originally published in Towards AI on Medium, where people are continuing the conversation by highlighting and responding to this story.

── more in #robotics 4 stories Β· sorted by recency
── more on @physical intelligence 3 stories trending now
sponsored brought to you by zahid.host 4,200+ EU-deployed projects
reading about agents? ship yours in a single git push.

Run your AI side-project on zahid.host

EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain β€” perfect for shipping the agent you just read about.

$git push zahid main
β†’ Live at https://your-agent.zahid.host βœ“
Get free account β†’ Pricing
from €0/mo Β· no card required
LIVE [news/demystifying-physica…] indexed:0 read:6min 2026-11-21 Β· β€”