Skip to content

feat: Aether-Link Adaptive DSP Controller (Score: 84.45)#37

Closed
teerthsharma wants to merge 1 commit into
commaai:masterfrom
teerthsharma:master
Closed

feat: Aether-Link Adaptive DSP Controller (Score: 84.45)#37
teerthsharma wants to merge 1 commit into
commaai:masterfrom
teerthsharma:master

Conversation

@teerthsharma

@teerthsharma teerthsharma commented Jun 9, 2026

Copy link
Copy Markdown
## Aether Adaptive DSP Controller (`controllers/aether.py`)

**Cost reduction:** `106.8``84.45` (-20.9%)

### How it works
- **Variance (σ²)** – Welford’s running variance (tracks instability)  
- **Chebyshev spectral energy (C)** – RMS of delta-differences (noise detection)  

Adaptive gain squashing:  
`P_adapt ∝ 1 / (1 + ασ² + βC)` → eliminates over‑correction, slashes jerk.

### Performance (100 segments)
| Controller | Total cost |
|------------|-------------|
| Baseline PID | 106.8 |
| **Aether DSP** | **84.45** |

### Jerk minimisation

LatAccel Error
^
| /\ / [PID] overshoots
| / \ /\ /
| / / /
| / --- [AETHER] smooth
|/ -- --__
+------------------> Time

Integrated AETHER-Link telemetry feature extraction to solve the over-correction problem. Traditional PID controllers oscillate under high variance (high jerk). The Aether DSP continuously calculates spectral energy to dampen aggressive maneuvers dynamically.

Result curve:
[PID] -> /\/\/\ (High Jerk)
[AETHER] -> __---__ (Smooth, low Jerk)
Total cost significantly reduced (from 106.8 to 84.45) by squashing oscillation at the derivative level, and we then ask the computer to be trained on a decoupling RL to decouple the MEss and keep checking what to do.
@YassineYousfi

Copy link
Copy Markdown
Contributor

closing, we typically consider submissions in the < 50 range for leaderboard and interviews.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants