Task 2: add Bio Player (pointwise-activation ESN with colour cue) — 14.78 ± 0.20

[GuanchunLi]

Summary

Adds braincraft/env2_player_bio.py, a handcrafted pointwise-activation Echo State Network controller for Task 2, together with a full model description in braincraft/env2_player_bio.md.

No learning / no training loop — the weight matrices are fixed at build time and the single yield in bio_player() returns the frozen model.

Approach

Extends the env1 bio player (same pointwise-ESN shape: every hidden activation is a scalar function of its own preactivation, all cross-neuron logic in W / Win) with the Task 2 colour cue. The controller combines:

• a reflex wall-follower (proximity / safety / hit taps),
• an initial-heading correction over a six-step seed window that cancels the ±5° start-direction perturbation,
• a rising-edge energy-reward detector that arms the shortcut,
• a heading-gated corridor shortcut triggered only when pos_x hits ±drift_offset with a horizontal heading,
• a blue-evidence circuit — per-ray blue-bump detectors feeding a signed integrator (evidence) with left/right sums; a ±color_evidence_thr threshold latches a signed front-block state (fs_pos / fs_neg),
• a signed front-block (front_block_pos, front_block_neg) driven by the two centre proximity taps and gated by the latched sign, so a frontal obstacle steers toward the open (blue) side rather than a fixed direction.

Readout has nine non-zero Wout entries; see env2_player_bio.md for the full activation library, slot layout, wiring equations, and gains.

Assumed performance

Running python braincraft/env2_player_bio.py (seed 12345, 10 runs):