CLI-PPO: A Memory-Augmented PPO Framework with Intrinsic Curiosity and Curriculum Learning for Mapless Multi-Obstacle Navigation

. 2026 Sep 25;26(19):6098. doi: 10.3390/s26196098

Algorithm 1 Adaptive Curriculum Learning

Require: Environment $Env$ , success threshold $ρ$ , buffer size N, curriculum levels L
Ensure: Curriculum level l, environment boundary Env.boundary

1:
Initialize $l \leftarrow 0$ , $τ \leftarrow 0$ , $B \leftarrow \emptyset$
2:
Set $ρ \leftarrow 0.7$ , $N \leftarrow 50$ , $L \leftarrow 10$ , $τ_{\max} \leftarrow 100$
3:
$α \leftarrow linspace (0.3, 1.0, L + 1)$
4:
Env.boundary $\leftarrow α [l]$
5:
for $s t e p = 1 : TotalTimesteps$ do
6:
if $τ > 0$ then
7:
$τ \leftarrow τ - 1$
8:
Execute_PPO_Step()
9:
continue
10:
end if
11:
$(d o n e, i s_s u c c e s s) \leftarrow$ Execute_PPO_Step()
12:
if $d o n e$ then
13:
Add( $B, i s_s u c c e s s$ )
14:
end if
15:
if $S i z e (B) \geq N$ then
16:
$R_{s u c c e s s} \leftarrow M e a n (B)$
17:
if $R_{s u c c e s s} \geq ρ$ and $l < L$ then
18:
$l \leftarrow l + 1$
19:
Env.boundary $\leftarrow α [l]$
20:
Clear( $B$ )
21:
$τ \leftarrow τ_{\max}$
22:
end if
23:
end if
24:
end for
25:
return l, Env.boundary