PROJECT FILM / ZONGO MAQUTU

A learning agent: early failure to later routes

Name: A learning agent: early failure to later routes
Uploaded: 2026-10-01T22:30:22+02:00
Duration: 12 s
Description: Compare episodes 1, 503 and 998 from an actual 1,000-episode learning run with seven clustered sink obstacles. An instrumented replay of the original application’s learning rule. It uses a scalar value table and is not a conventional state–action Q-table. The Field Note documents that distinction and the capture repairs.

Compare episodes 1, 503 and 998 from an actual 1,000-episode learning run with seven clustered sink obstacles.

12 seconds · actual app, selected episodes 1 → 503 → 998 · original learning rule, instrumented replay. · Silent demonstration.

What the film shows

The first selected episode shows an early failed journey.
Later selected episodes show successful routes around seven clustered sink states.
Episode numbers and move counts make the progression visible.

Behind the demonstration

An instrumented replay of the original application’s learning rule. It uses a scalar value table and is not a conventional state–action Q-table. The Field Note documents that distinction and the capture repairs.

Built and documented by Zongo Maqutu. Read the Field Note for the implementation, architecture and limitations.

Read Field Note Launch Project Source