平台跳跃与街机反应
Platformers & arcade reflex
Progress through levels with timed inputs from pixels.
20 条本项目编写的提案 · 分类 v1.1 · 新增基准运行 0 次。下方验证方式与部分得分规则是设计，不是执行结果。
- 典型应用
- PyBoy · Chrome Dino · arcade emulators
- 小类 GUI 需求标签
G3· 仅 GUI
任务提案 · 20
展开查看完整原文与建议检查方式；两种界面均保留提案英文原文。
R13.01.01.01-T01 Clear Super Mario Land world 1-1 in PyBoy from pixels
原始指令
A PyBoy window shows Super Mario Land (harness-supplied ROM) at the start of world 1-1 with 3 lives. Using only the keyboard bindings shown in controls.txt, reach the end-of-level gate. The game runs in real time; no pausing or savestates.
- 平台
mac· macOSwin· Windowslin· Linux- 应用
- PyBoy
- 交互表面
desk· 桌面 GUI- GUI 需求 · 任务设定
G3· 仅 GUI- GUI 需求 · 普通权限
G2· GUI 必要- 交互原语
rt· 实时控制key· 快捷键vis· 视觉判断- 工作产物
WP11· 游戏 / 仿真状态- 条件
- 未声明
- 提案难度
- 3 / 5
建议验证方式
Harness reads RAM each frame: pass when the level-complete flag is set. The agent sees only frames and sends OS key events; PyBoy's Python API, savestate hotkeys and speed-up are disabled, and the input log is replayed deterministically to confirm the run.
建议部分得分
50% scaled by x-progress over the first half of the level (RAM scroll); 50% scaled over the second half, full credit on level-complete flag
R13.01.01.01-T02 Reach 500 points in the offline dinosaur runner
原始指令
A local copy of the Chrome offline dinosaur game is open in the browser at the start screen. Press space to start and reach a score of at least 500 in a single run by jumping cacti and ducking birds.
- 平台
web· 任意系统（网页）- 应用
- Chrome Dino
- 交互表面
brow· 浏览器- GUI 需求 · 任务设定
G3· 仅 GUI- GUI 需求 · 普通权限
G2· GUI 必要- 交互原语
rt· 实时控制key· 快捷键vis· 视觉判断- 工作产物
WP11· 游戏 / 仿真状态- 条件
- 未声明
- 提案难度
- 2 / 5
建议验证方式
Injected game-state bridge reports Runner distance/score; pass at ≥ 500 in one run. Devtools are disabled and Runner object properties are frozen, so speed or collision cannot be patched; only trusted key events are accepted.
建议部分得分
50% scaled by best single-run score up to 250; 50% scaled from 250 to 500
R13.01.01.01-T03 Reach height 2,000 in a Doodle Jump-style web game
原始指令
A self-hosted Doodle Jump clone (fixed seed) is open. Steer with the arrow keys so the character keeps landing on platforms and reaches a height of 2,000 within 3 attempts. Moving and breaking platforms appear above 800.
- 平台
web· 任意系统（网页）- 应用
- arcade emulators
- 交互表面
brow· 浏览器- GUI 需求 · 任务设定
G3· 仅 GUI- GUI 需求 · 普通权限
G2· GUI 必要- 交互原语
rt· 实时控制key· 快捷键vis· 视觉判断- 工作产物
WP11· 游戏 / 仿真状态- 条件
- 未声明
- 提案难度
- 3 / 5
建议验证方式
JS bridge logs max height per attempt and the attempt count; pass when any of the first 3 attempts reaches 2,000. Game objects are closure-scoped and DevTools are disabled by browser policy, so state cannot be patched by script; key events must be trusted (isTrusted).
建议部分得分
50% scaled by best height up to 1,000; 50% scaled from 1,000 to 2,000
R13.01.01.01-T04 Finish Super Mario Bros 1-1 with at least ten coins
原始指令
An emulator window (FCEUX, harness-supplied ROM) shows Super Mario Bros at world 1-1. Finish the level by touching the flagpole with at least 10 coins collected in that level. Use the keyboard controls in controls.txt; savestates and rewind are disabled.
- 平台
mac· macOSwin· Windowslin· Linux- 应用
- arcade emulators
- 交互表面
desk· 桌面 GUI- GUI 需求 · 任务设定
G3· 仅 GUI- GUI 需求 · 普通权限
G2· GUI 必要- 交互原语
rt· 实时控制key· 快捷键vis· 视觉判断- 工作产物
WP11· 游戏 / 仿真状态- 条件
- 未声明
- 提案难度
- 4 / 5
建议验证方式
RAM read by the harness: coin counter ≥ 10 at the flagpole event and level-complete for 1-1. Lua scripting and memory tools are disabled in the emulator build; input log replay confirms the run.
建议部分得分
50% scaled by x-progress; 20% flagpole reached; 30% scaled by coins up to 10
R13.01.01.01-T05 Pass ten pipes in a Flappy Bird-style web game
原始指令
A self-hosted Flappy Bird clone with a fixed pipe seed is open. Tap space or click to flap and pass 10 pipes in one run.
- 平台
web· 任意系统（网页）- 应用
- arcade emulators
- 交互表面
brow· 浏览器- GUI 需求 · 任务设定
G3· 仅 GUI- GUI 需求 · 普通权限
G2· GUI 必要- 交互原语
rt· 实时控制key· 快捷键vis· 视觉判断- 工作产物
WP11· 游戏 / 仿真状态- 条件
- 未声明
- 提案难度
- 2 / 5
建议验证方式
JS bridge reports pipes passed per run; pass at ≥ 10. Game loop runs at a fixed 60 Hz and does not pause while the agent thinks. Game objects are closure-scoped and DevTools are disabled by browser policy, so state cannot be patched by script; only trusted (isTrusted) input is accepted.
建议部分得分
50% scaled by best run up to 5 pipes; 50% scaled from 5 to 10 pipes
R13.01.01.01-T06 Clear twenty bricks in a Breakout web game within three lives
原始指令
A self-hosted Breakout game (fixed seed) is open. Move the paddle with the mouse or arrow keys and clear at least 20 bricks before losing 3 lives.
- 平台
web· 任意系统（网页）- 应用
- arcade emulators
- 交互表面
brow· 浏览器- GUI 需求 · 任务设定
G3· 仅 GUI- GUI 需求 · 普通权限
G2· GUI 必要- 交互原语
rt· 实时控制vis· 视觉判断drag· 拖放- 工作产物
WP11· 游戏 / 仿真状态- 条件
- 未声明
- 提案难度
- 3 / 5
建议验证方式
JS bridge: bricks cleared when lives reach 0 or at 20. Game objects are closure-scoped and DevTools are disabled by browser policy, so state cannot be patched by script; the paddle accepts only trusted (isTrusted) pointer/key events.
建议部分得分
50% scaled by bricks cleared up to 10; 50% scaled from 10 to 20 bricks
R13.01.01.01-T07 Clear all pellets on level 1 of a Pac-Man web clone
原始指令
A self-hosted Pac-Man clone (deterministic ghost AI) is open with 3 lives. Eat every pellet on level 1 so the maze advances to level 2.
- 平台
web· 任意系统（网页）- 应用
- arcade emulators
- 交互表面
brow· 浏览器- GUI 需求 · 任务设定
G3· 仅 GUI- GUI 需求 · 普通权限
G2· GUI 必要- 交互原语
rt· 实时控制key· 快捷键vis· 视觉判断- 工作产物
WP11· 游戏 / 仿真状态- 条件
- 未声明
- 提案难度
- 4 / 5
建议验证方式
JS bridge: pellet count 0 and level = 2 before lives reach 0. Ghost AI is deterministic so the trajectory replays exactly. Game objects are closure-scoped and DevTools are disabled by browser policy, so state cannot be patched by script; only trusted (isTrusted) keys are accepted.
建议部分得分
50% scaled by the first half of pellets eaten; 50% scaled by the remaining pellets, full credit at level 2
R13.01.01.01-T08 Grow the snake to length 25 in a snake web game
原始指令
A self-hosted Snake game (Google Snake-style, standard speed, fixed apple seed) is open. Steer with the arrow keys to reach a snake length of 25 without hitting a wall or yourself.
- 平台
web· 任意系统（网页）- 应用
- arcade emulators
- 交互表面
brow· 浏览器- GUI 需求 · 任务设定
G3· 仅 GUI- GUI 需求 · 普通权限
G2· GUI 必要- 交互原语
rt· 实时控制key· 快捷键vis· 视觉判断- 工作产物
WP11· 游戏 / 仿真状态- 条件
- 未声明
- 提案难度
- 3 / 5
建议验证方式
JS bridge: max length in one run; pass at 25. Speed is fixed and the game does not pause between frames. Game objects are closure-scoped and DevTools are disabled by browser policy, so state cannot be patched by script; only trusted (isTrusted) key events are accepted.
建议部分得分
50% scaled by max length from 4 to 15; 50% scaled from 15 to 25
R13.01.01.01-T09 Clear 40 lines in a Tetris sprint under five minutes
原始指令
A self-hosted Tetris (guideline rules, 7-bag with fixed seed, hold and hard drop enabled) is open in Sprint mode. Clear 40 lines in under 5:00 without topping out.
- 平台
web· 任意系统（网页）- 应用
- arcade emulators
- 交互表面
brow· 浏览器- GUI 需求 · 任务设定
G3· 仅 GUI- GUI 需求 · 普通权限
G2· GUI 必要- 交互原语
rt· 实时控制key· 快捷键vis· 视觉判断- 工作产物
WP11· 游戏 / 仿真状态- 条件
- 未声明
- 提案难度
- 4 / 5
建议验证方式
Game reports lines cleared and elapsed time; pass at 40 lines ≤ 300 s. Gravity is time-based, so slow decisions cost time and cause lock-outs. Game objects are closure-scoped and DevTools are disabled by browser policy, so state cannot be patched by script; only trusted (isTrusted) keys are accepted.
建议部分得分
70% scaled by lines cleared up to 40; 30% if finished under 5:00
R13.01.01.01-T10 Defeat the first boss of Kirby's Dream Land in PyBoy
原始指令
PyBoy shows Kirby's Dream Land (harness-supplied ROM) at the start of Green Greens. Clear the stage and defeat Whispy Woods using inhale-and-spit on the apples. No savestates.
- 平台
mac· macOSwin· Windowslin· Linux- 应用
- PyBoy
- 交互表面
desk· 桌面 GUI- GUI 需求 · 任务设定
G3· 仅 GUI- GUI 需求 · 普通权限
G2· GUI 必要- 交互原语
rt· 实时控制key· 快捷键vis· 视觉判断- 工作产物
WP11· 游戏 / 仿真状态- 条件
- 未声明
- 提案难度
- 4 / 5
建议验证方式
RAM: stage index advances past Green Greens with the boss-defeated flag. PyBoy API and savestates disabled; inputs replayed to confirm.
建议部分得分
40% scaled by stage progress; 30% boss reached; 30% boss defeated
R13.01.01.01-T11 Clear the first Space Invaders wave without losing a life
原始指令
MAME (harness-supplied Space Invaders set) is running at the title screen with credits inserted. Start a 1-player game and destroy all 55 invaders of wave 1 without losing a life.
- 平台
mac· macOSwin· Windowslin· Linux- 应用
- arcade emulators
- 交互表面
desk· 桌面 GUI- GUI 需求 · 任务设定
G3· 仅 GUI- GUI 需求 · 普通权限
G3· 仅 GUI- 交互原语
rt· 实时控制key· 快捷键vis· 视觉判断- 工作产物
WP11· 游戏 / 仿真状态- 条件
- 未声明
- 提案难度
- 4 / 5
建议验证方式
MAME RAM read by the harness (debugger disabled for the agent): wave counter advances with lives unchanged. Cheats and save states are disabled.
建议部分得分
70% scaled by invaders destroyed; 30% no life lost
R13.01.01.01-T12 Reach the 500 m summit stage in a PICO-8 precision platformer
原始指令
A local PICO-8 web export of Celeste Classic is open. Climb from 100 m to at least 500 m (stage 5) with no more than 20 deaths, using dash and wall-jump.
- 平台
web· 任意系统（网页）- 应用
- arcade emulators
- 交互表面
brow· 浏览器- GUI 需求 · 任务设定
G3· 仅 GUI- GUI 需求 · 普通权限
G2· GUI 必要- 交互原语
rt· 实时控制key· 快捷键vis· 视觉判断- 工作产物
WP11· 游戏 / 仿真状态- 条件
- 未声明
- 提案难度
- 5 / 5
建议验证方式
PICO-8 GPIO/memory bridge exposes level index and death counter to the harness only; pass when level ≥ 5 with deaths ≤ 20. DevTools are disabled by browser policy, so cartridge memory cannot be poked from the page; only trusted (isTrusted) keys are accepted and the cartridge's level-skip cheat is removed.
建议部分得分
80% scaled by levels cleared (one fifth per level); 20% within death limit
R13.01.01.01-T13 Finish Super Mario Bros 1-1 with at least 330 time remaining
原始指令
In FCEUX (harness-supplied ROM), complete world 1-1 of Super Mario Bros with the in-game timer showing 330 or more at the flagpole. Pipe shortcuts are allowed.
- 平台
mac· macOSwin· Windowslin· Linux- 应用
- arcade emulators
- 交互表面
desk· 桌面 GUI- GUI 需求 · 任务设定
G3· 仅 GUI- GUI 需求 · 普通权限
G2· GUI 必要- 交互原语
rt· 实时控制key· 快捷键vis· 视觉判断- 工作产物
WP11· 游戏 / 仿真状态- 条件
- 未声明
- 提案难度
- 5 / 5
建议验证方式
RAM: timer value at flagpole ≥ 330 and level-complete. Rewind, savestates and Lua disabled; input log replayed.
建议部分得分
40% level completed; 60% scaled by timer from 250 to 330
R13.01.01.01-T14 Achieve a 70% hit ratio on Galaga stage 1
原始指令
In MAME (harness-supplied Galaga set), play stage 1 and destroy every enemy of the stage with a shot hit ratio of at least 70%, without losing a ship.
- 平台
mac· macOSwin· Windowslin· Linux- 应用
- arcade emulators
- 交互表面
desk· 桌面 GUI- GUI 需求 · 任务设定
G3· 仅 GUI- GUI 需求 · 普通权限
G2· GUI 必要- 交互原语
rt· 实时控制key· 快捷键vis· 视觉判断- 工作产物
WP11· 游戏 / 仿真状态- 条件
- 未声明
- 提案难度
- 4 / 5
建议验证方式
RAM counters for shots fired, hits, stage and ships read by the harness: stage 2 reached, hits/shots ≥ 0.70, no ship lost. MAME's debugger, Lua console, cheats and save states are disabled in the harness build, and the input log is replayed to confirm the run.
建议部分得分
40% stage cleared; 40% scaled by hit ratio from 0.4 to 0.7; 20% no ship lost
R13.01.01.01-T15 Run 1,000 m in a three-lane endless runner
原始指令
A self-hosted 3D endless runner (Temple Run-style, fixed obstacle seed) is open in the browser. Swipe-equivalent arrow keys change lane, jump and slide. Reach 1,000 m in a single run.
- 平台
web· 任意系统（网页）- 应用
- arcade emulators
- 交互表面
brow· 浏览器- GUI 需求 · 任务设定
G3· 仅 GUI- GUI 需求 · 普通权限
G2· GUI 必要- 交互原语
rt· 实时控制key· 快捷键vis· 视觉判断- 工作产物
WP11· 游戏 / 仿真状态- 条件
- 未声明
- 提案难度
- 3 / 5
建议验证方式
JS bridge reports distance per run; pass at ≥ 1,000 m. Speed ramps over time. Game objects are closure-scoped and DevTools are disabled by browser policy, so state cannot be patched by script; only trusted (isTrusted) input is accepted.
建议部分得分
50% scaled by best distance up to 500 m; 50% scaled from 500 to 1,000 m
R13.01.01.01-T16 Get five frogs home in a Frogger web clone
原始指令
A self-hosted Frogger clone (fixed traffic pattern) is open. Guide frogs across the road and river into all 5 home bays before the round timer and lives run out.
- 平台
web· 任意系统（网页）- 应用
- arcade emulators
- 交互表面
brow· 浏览器- GUI 需求 · 任务设定
G3· 仅 GUI- GUI 需求 · 普通权限
G2· GUI 必要- 交互原语
rt· 实时控制key· 快捷键vis· 视觉判断- 工作产物
WP11· 游戏 / 仿真状态- 条件
- 未声明
- 提案难度
- 3 / 5
建议验证方式
JS bridge: occupied home bays and lives; pass when all 5 bays are filled in one game. Game objects are closure-scoped and DevTools are disabled by browser policy, so state cannot be patched by script; only trusted (isTrusted) key events are accepted.
建议部分得分
20% first bay filled; 20% second bay; 20% third bay; 40% fourth and fifth bays
R13.01.01.01-T17 Complete level 1 of a Geometry Dash-style rhythm platformer
原始指令
A self-hosted Geometry Dash-style clone shows level 1 (auto-scrolling, about 70 s, synced to music). Jump spikes and gaps with a single key and reach the end. Practice mode checkpoints are disabled.
- 平台
web· 任意系统（网页）- 应用
- arcade emulators
- 交互表面
brow· 浏览器- GUI 需求 · 任务设定
G3· 仅 GUI- GUI 需求 · 普通权限
G2· GUI 必要- 交互原语
rt· 实时控制key· 快捷键av· 音视频时序- 工作产物
WP11· 游戏 / 仿真状态- 条件
- 未声明
- 提案难度
- 5 / 5
建议验证方式
JS bridge: best completion percentage over attempts and the completion flag; full credit on completion within 60 minutes of wall-clock time. Game objects are closure-scoped and DevTools are disabled by browser policy, so state cannot be patched by script; only trusted (isTrusted) key events are accepted.
建议部分得分
50% scaled by best completion up to the halfway point; 50% scaled from halfway to the level end
R13.01.01.01-T18 Score one point against the CPU in a Pong web game
原始指令
A self-hosted Pong game is open with a slow CPU opponent. Move your paddle with the mouse and score one point against the CPU before it scores 3.
- 平台
web· 任意系统（网页）- 应用
- arcade emulators
- 交互表面
brow· 浏览器- GUI 需求 · 任务设定
G3· 仅 GUI- GUI 需求 · 普通权限
G2· GUI 必要- 交互原语
rt· 实时控制drag· 拖放vis· 视觉判断- 工作产物
WP11· 游戏 / 仿真状态- 条件
- 未声明
- 提案难度
- 1 / 5
建议验证方式
JS bridge: player score ≥ 1 while CPU score < 3. Game objects are closure-scoped and DevTools are disabled by browser policy, so state cannot be patched by script; the paddle accepts trusted (isTrusted) pointer events only.
建议部分得分
30% returns the ball at least 3 times; 70% scores a point
R13.01.01.01-T19 Reach level 5 in Game Boy Tetris A-type in PyBoy
原始指令
PyBoy shows Tetris (harness-supplied ROM) at the game-type menu. Start A-type at level 0 and play until the level indicator shows 5 (50 lines), without topping out.
- 平台
mac· macOSwin· Windowslin· Linux- 应用
- PyBoy
- 交互表面
desk· 桌面 GUI- GUI 需求 · 任务设定
G3· 仅 GUI- GUI 需求 · 普通权限
AP· API 足够- 交互原语
rt· 实时控制key· 快捷键vis· 视觉判断- 工作产物
WP11· 游戏 / 仿真状态- 条件
long· 长流程- 提案难度
- 3 / 5
建议验证方式
RAM: level byte ≥ 5 during a single game. PyBoy API, savestates and speed control disabled; input replay confirms.
建议部分得分
50% scaled by lines cleared up to 25; 50% scaled from 25 to 50 lines
R13.01.01.01-T20 Reach 300 points in the dinosaur runner on Android with taps
原始指令
On the Android emulator, a local web copy of the dinosaur runner is open in Chrome in portrait mode. Start with a tap and reach a score of at least 300 using taps to jump; the run speeds up over time.
- 平台
and· Android- 应用
- Chrome Dino
- 交互表面
mob· 移动触控- GUI 需求 · 任务设定
G3· 仅 GUI- GUI 需求 · 普通权限
G2· GUI 必要- 交互原语
touch· 触控手势rt· 实时控制vis· 视觉判断- 工作产物
WP11· 游戏 / 仿真状态- 条件
- 未声明
- 提案难度
- 3 / 5
建议验证方式
JS bridge through adb-forwarded devtools reports best score; pass at ≥ 300. That DevTools port is reachable only by the harness, not the agent, and game objects are closure-scoped, so state cannot be patched by script; input must arrive as touch events from the emulator.
建议部分得分
50% scaled by best score up to 150; 50% scaled from 150 to 300
已有研究映射
以下沿用研究目录中的基准与小类映射，不是逐题标注、执行记录或产品背书。