Observed arrival · 2026-11-18
Total Kinematics puts frontier models in a robot’s shoes
A humanoid-robot project pairs its TK1 platform with KineBench, a benchmark for testing how frontier models handle robotics tasks.
○Open
⊠Login
$Paid
†Ads
✦Pretty
●Pro
◎Niche
◉Human
⚑Risk
ƒJS
- For
- Robotics researchers and embodied-AI evaluators
- Worth noticing
- The homepage lists six KineBench task types, with simulation scores based on ground truth rather than model self-report.
Field notes
The TK1 section pairs hardware details with a displayed tool-call sequence: the model looks through a depth camera, locates the bottle, walks, picks it up, and speaks. KineBench describes tasks as scenes plus instructions, with tool calls over MCP and scores determined by simulation ground truth. The project says the current platform is a modified Berkeley Humanoid Lite and identifies a bespoke TK2 as the next step.
Observed signals
Read the marks
Editorial observations of this landing page, not a rating.
○OpenPublic substance visible
✦PrettyNotable craft visible
●ProPolished or operationally mature
◎NicheUnusually specific use
One card from the complete issue