Goal-conditioned RL — setup.

medium

Answer

${\pi}(a\; \mid \;s, \;g)$
$Reward r(s, a, g) = f(distance to g) or -1 until g reached.$
$Enables one policy for many tasks.$
$Combined with HER (relabel failed attempts) for sample efficiency.$
$Foundation of universal value function approximation (UVFA, Schaul 2015).$
$Modern uses: robot manipulation reaching diverse target poses, hierarchical HIRO planning, generalist RL agents.$

Check yourself — multiple choice

Random
${\pi}(a\; \mid \;s, \;g)$ conditioned on desired goal; reward from distance to g; one policy for many tasks; HER for sample efficiency; UVFA foundation; modern generalist agents
Same as vanilla
Not real

Goal-conditioned RL: ${\pi}(a\; \mid \;s, \;g)$ ; UVFA + HER foundation.

#policy-methods

214 interview questions in this topic.