AI and Robot Learning
Domain Randomisation: The Trick That Makes Simulation Work
Instead of modelling reality accurately, randomise every parameter you cannot measure so that the real world becomes one sample from the...
Section 06
AI and Robot Learning
Policies, datasets and the honest state of learned control on real hardware.
AI and Robot Learning
Imitation Learning vs Reinforcement Learning for Robots
Use imitation learning when a human can demonstrate the task, which is almost always in manipulation. Use reinforcement learning when...
AI and Robot Learning
How Much Compute Does On-Robot Inference Need?
A perception stack with a learned policy typically needs 20 to 200 TOPS of accelerator throughput and draws 15 to 120 W, which on a battery...
AI and Robot Learning
Open Robot Datasets Compared: Size, Embodiments, Licence
Check three properties before anything else: the action space, the observation setup and the licence. A dataset recorded in joint space on...
AI and Robot Learning
Why Reinforcement Learning Rarely Reaches the Factory
Five obstacles, and the decisive ones are not algorithmic. A factory needs predictable behaviour, certifiable safety, diagnosable faults,...
AI and Robot Learning
Robot Foundation Models: What Works Today
Generalisation across objects works, generalisation across robots works with the right action representation, and generalisation across...
AI and Robot Learning
Robot Grasping: What Success Rate Is State of the Art?
Reported grasp success on unknown objects sits around 85 % to 95 % in controlled evaluations, and that is genuinely good research. It is...
AI and Robot Learning
How Much Data Does a Robot Policy Need?
For a fixed task with one object in one place, 50 to 150 demonstrations often produce a working policy. Add object variety, pose variation,...
AI and Robot Learning
The Sim-to-Real Gap: Why Policies Fail on Hardware
The gap is not mysterious, it is a list. Friction, actuator dynamics, timing, sensor noise structure, contact geometry and mass...
AI and Robot Learning
Vision Language Action Models: What They Actually Do
A vision language action model takes an image and a natural language instruction and outputs robot actions directly, usually end effector...