Robotics Weekly

A Weekly Editorial on Robotics & Artificial Intelligence

Topic Hub

Physical AI

How vision-language-action (VLA) models and foundation models translate into the physical world, where the failure modes are different from chat and code.

What we track

  • · VLA model architectures
  • · Sim-to-real transfer
  • · Embodied AI benchmarks
A whiteboard diagram contrasting rigid robotic arm paths with adaptive robot reasoning

vla-models

Defining the Stack: Why "Levels of Autonomy" is Dead

BCG and Deloitte just retired the automotive industry's L1–L5 framework for robots. The new benchmark is causal reasoning, not scripted precision.

No. 7 of 7 · Priya Raman

A robotic hand fitted with tactile sensors gripping a fragile object

vla-models

The Tactile Missing Link: Why Vision and Language Aren't Enough for Embodied AI

RobOmni, the industry's first tactile-sensing benchmark from Daimon Robotics and Galbot, exposes the grounding bottleneck inside vision-language-action models.

No. 1 of 2 · Maya Chen

A robot arm wired directly into an edge-compute control board on a factory bench

infrastructure

Beyond the Foundation Model: Why the Future of Robotics Belongs to Native Hardware

GENISOM AI's new mass-produced robotics platform bypasses the cloud-wrapped control stack entirely, betting that deployability, not model scale, is now the real differentiator.

No. 2 of 2 · Daniel Osei

Advertisement

Intelligence, Embodied. Safety, Guaranteed.

Xolver gives robot arms a foundation model that sees and adapts in real time — and an enforcement layer that makes unsafe motion architecturally impossible. Retrofit existing hardware or build on the VLA model SDK. Either way, autonomy you can audit, not just hope works.

Explore xolver.ai →