State vector additions #21

New Issue

2026-08-17T16:58:12+02:00

SirStone commented

2026-08-17 16:58:12 +02:00

Blocked by: Output space design for goto/aimTo

Parent map: #18

Question

What new inputs should be added to the state vector to close the feedback loop on in-flight commands?

Candidates:

Remaining distance to goto target (scalar)
Remaining gun angle to aimTo target (scalar)
Bearing to goto target relative to current heading?
Time (ticks) since current command was issued?
Current goto target (x, y) — or is remaining distance sufficient?

How should these be normalized? Current state vector is 42-dim — what's the new size?

**Blocked by:** [Output space design for goto/aimTo](#19) Parent map: #18 ## Question What new inputs should be added to the state vector to close the feedback loop on in-flight commands? Candidates: - Remaining distance to goto target (scalar) - Remaining gun angle to aimTo target (scalar) - Bearing to goto target relative to current heading? - Time (ticks) since current command was issued? - Current goto target (x, y) — or is remaining distance sufficient? How should these be normalized? Current state vector is 42-dim — what's the new size?

SirStone added the wayfinder:grilling label 2026-08-17 16:58:12 +02:00

SirStone commented

2026-08-17 17:30:30 +02:00

Resolution

2 new state inputs (42 → 44 dims):

New input	Normalization
Remaining distance to goto target	÷ battlefield diagonal → [0, 1]
Remaining gun angle to aimTo target	÷ 180° → [0, 1]

Decided against adding:

Previous goto target (x, y) — the network sees its own position + remaining distance each tick, which is sufficient signal to learn consistency. Feeding back the previous target adds complexity the network must learn to interpret.
Bearing to goto target — derivable from position + heading already in state.
Ticks since command — redundant with remaining distance.
Current aimTo target — remaining angle is sufficient, gun turns are fast.

## Resolution **2 new state inputs (42 → 44 dims):** | New input | Normalization | |-----------|--------------| | Remaining distance to goto target | ÷ battlefield diagonal → [0, 1] | | Remaining gun angle to aimTo target | ÷ 180° → [0, 1] | **Decided against adding:** - Previous goto target (x, y) — the network sees its own position + remaining distance each tick, which is sufficient signal to learn consistency. Feeding back the previous target adds complexity the network must learn to interpret. - Bearing to goto target — derivable from position + heading already in state. - Ticks since command — redundant with remaining distance. - Current aimTo target — remaining angle is sufficient, gun turns are fast.

SirStone closed this issue

2026-08-17 17:30:39 +02:00

SirStone referenced this issue

2026-08-17 17:30:55 +02:00

Command abstraction layer for PPO action space #18

SirStone referenced this issue

2026-08-17 18:51:59 +02:00

PRD: Command abstraction layer for PPO action space #24

SirStone referenced this issue

2026-08-18 16:32:24 +02:00

Command abstraction layer for PPO action space #18

Sign in to join this conversation.