Milestone Roadmap — From Static Scene Demo to Real-Time Path Tracing

The Himalaya renderer follows a three-milestone progressive enhancement strategy, each delivering a self-contained visual quality plateau. This roadmap traces the full arc from a static scene demo with basic PBR lighting through to a real-time path-traced rendering mode comparable to commercial game engines. Every milestone has a clear "look and feel" target, a bounded feature set, and explicitly declared limitations that are resolved in subsequent milestones.

Sources: milestone-1.md, milestone-2.md, milestone-3.md

The Big Picture

Himalaya's roadmap is organized around a single architectural conviction: use offline quality to solve real-time problems. Baked Lightmaps and Reflection Probes deliver photorealistic indirect lighting at nearly zero runtime cost. Screen-space effects (GTAO, Contact Shadows, future SSR/SSGI) provide the layer of dynamic detail that baking cannot capture. Path tracing serves as both an offline baking tool and a real-time rendering mode. This layered approach means each milestone builds upon — rather than replaces — the previous one.

Sources: requirements-and-philosophy.md, technical-decisions.md, milestone-future.md

Guiding Principles Behind the Roadmap

Each milestone's content is shaped by nine design principles documented in the project's requirements specification. The three most impactful on the roadmap are:

• Progressive enhancement — every technology follows a Pass 1 → Pass 2 → Pass 3 evolution. Forward rendering goes brute-force → Tiled → Clustered. Shadows go basic Shadow Map → CSM → CSM+PCSS. GI goes ambient → baked Lightmap → SSGI. This means each milestone "locks in" a quality plateau while keeping the upgrade path open.

• Performance sweet spot — technologies are chosen at the point where implementation cost is proportional to visual or performance gain. GPU-Driven Rendering, Virtual Shadow Maps, and Mesh Shaders are explicitly excluded because the project's scene scale does not justify their complexity.

• Infrastructure reuse — a single investment in temporal filtering (Milestone 1 Phase 5) serves GTAO, future SSR, and SSGI. The depth PrePass serves forward rendering, all screen-space effects, and hardware occlusion queries. The RT infrastructure (Phase 6) serves the baking tool, the reference viewer, and the full real-time PT mode.

Sources: requirements-and-philosophy.md, technical-decisions.md

Milestone 1 — Static Scene Demo

Goal: A static scene with free camera movement, physically-based shading, multi-cascade shadows, screen-space ambient occlusion, baked indirect lighting, and a path-traced reference viewer. Indoor and outdoor scenes both supported.

Visual Target

Indoor scenes feature baked indirect lighting (Lightmap) with soft light bouncing through windows. Outdoor scenes have directional light shadows covering near-to-far distances (CSM), contact-hardening soft shadows (PCSS), and per-pixel contact shadows at object-ground contact points. All surfaces use Cook-Torrance PBR with split-sum IBL. A path-traced reference view with OIDN denoising provides ground-truth comparison.

Sources: milestone-1.md

Development Phases

Milestone 1 is divided into ten sequential phases, each producing a visible, verifiable result. The dependency chain ensures that no phase works on infrastructure for too long without producing something observable.

Sources: m1-development-order.md

Phase Dependency Graph

Sources: m1-development-order.md

Current Status (Phase 6 In Progress)

Phase 6 is the current active phase, implementing the RT infrastructure and path-traced reference viewer. It consists of 13 steps, with Steps 1–10 completed (115 tasks done, 42 remaining). The remaining work covers advanced PT sampling techniques:

Sources: m1-phase6.md, current-phase.md

Known Limitations

Milestone 1 intentionally scopes out several visual features. Each limitation has a clear resolution milestone:

Sources: milestone-1.md

Milestone 2 — Quality Leap + Real-Time Path Tracing

Goal: On top of M1's foundation, capture all low-hanging visual fruit and introduce a real-time path-traced rendering mode. The rasterized mode should achieve a quality level noticeably superior to M1, while the PT mode targets visual fidelity comparable to DOOM: The Dark Ages path-traced mode.

Visual Target

The rasterized mode gains: dynamic sky with day/night cycle (Bruneton atmospheric scattering replacing static Cubemap), physically correct aerial perspective (replacing height fog), screen-space reflections and indirect lighting, and FSR/DLSS upscaling with temporal stability. The PT mode provides real-time path tracing with reservoir-based spatiotemporal resampling.

Sources: milestone-2.md

Feature Breakdown by Effort

Sources: milestone-2.md

Real-Time Path Tracing Mode

The PT mode builds directly on M1's RT infrastructure (acceleration structures, RT pipeline abstraction, scene AS builder). Three key algorithms bring real-time path tracing to fruition:

Sources: milestone-2.md, technical-decisions.md

Hybrid RT Effects

In rasterized mode, RT selectively replaces only the techniques where rasterization has the most visible artifacts:

Sources: milestone-2.md

Milestone 3 — Dynamic Objects + Scale + PT Upgrade

Goal: Support dynamic objects (non-character), fill remaining visual gaps, provide performance tooling for larger scenes, and upgrade PT indirect light quality.

Milestone 3's three directions can be prioritized flexibly based on demo needs: showing dynamic objects, larger scenes, or indoor lighting atmosphere. This adaptability is intentional — the milestone delivers a toolkit rather than a fixed sequence.

Sources: milestone-3.md

Feature Matrix

Sources: milestone-3.md

Path Tracing Evolution Across Milestones

The PT rendering capability follows its own progressive enhancement arc, spanning all three milestones and beyond:

Sources: technical-decisions.md, milestone-future.md

Technology Replacement Chains

Several technologies are explicitly designed to be replaced, not accumulated. This keeps the rendering stack lean and avoids maintaining redundant systems:

Sources: requirements-and-philosophy.md, technical-decisions.md

What Is Explicitly Excluded

Not every rendering technology fits the project's design constraints. The following are documented as "will not implement" with clear reasoning:

Sources: milestone-future.md

Future Horizons

Beyond M3, the roadmap records technologies with explicit trigger conditions — they are not scheduled but are documented so architectural decisions today don't close doors:

• Neural rendering (NRC, DLSS Ray Reconstruction, Cooperative Vectors) — depends on Tensor Core availability and cross-vendor standardization
• ReSTIR PT — unifies ReSTIR DI + ReSTIR GI under the GRIS framework for full light transport path resampling
• Irradiance Probes + SDF GI — conditional on SDF infrastructure existing; marginal benefit over Lightmap + SSGI unless the scene has dynamic changes in off-screen areas
• HDR output — requires dual tonemapping parameters, UI brightness handling, and calibration interface
• Water rendering — most needs already covered by existing infrastructure (SSR, refraction, PBR Fresnel)

Sources: milestone-future.md

Key Infrastructure That Bridges Milestones

Several foundational systems are built once in M1 and serve multiple milestones:

Sources: m1-development-order.md, m1-design-decisions-core.md

Next reading: For the design philosophy driving these milestone boundaries, see Design Principles — Tradeoffs, Progressive Enhancement, and Sweet-Spot Engineering. For the detailed technical decisions underlying each feature, see the corresponding layer deep-dive pages in the catalog.