Tasks
Procedurally generated tasks across 6 categories. All tasks support 4 difficulty levels (easy/medium/hard/expert), multiple observation modes, and both sparse and dense rewards.
Browse tasks interactively — the best way to understand each task is to play it yourself in the webapp:
uv run python -m agentick.human.webapp # Opens http://localhost:5000
Or create one programmatically:
import agentick
env = agentick.make("GoToGoal-v0", difficulty="medium", render_mode="language", seed=42)
Task Catalog
Navigation (8 tasks)
| Task |
Description |
| GoToGoal-v0 |
Navigate to a visible goal in an open grid |
| MazeNavigation-v0 |
Solve procedurally generated mazes |
| ShortestPath-v0 |
Visit multiple goals in shortest-path order |
| DynamicObstacles-v0 |
Navigate while avoiding moving obstacles |
| CuriosityMaze-v0 |
Explore a maze to reach a coverage target |
| RecursiveRooms-v0 |
Navigate hierarchical room-in-room structures |
| TimingChallenge-v0 |
Pass through moving gates at precise moments |
| InstructionFollowing-v0 |
Navigate to annotated target, avoid distractors |
Planning (9 tasks)
| Task |
Description |
| SokobanPush-v0 |
Push boxes onto target positions (classic Sokoban) |
| KeyDoorPuzzle-v0 |
Collect color-coded keys to unlock matching doors; hard+ requires backtracking |
| BacktrackPuzzle-v0 |
Navigate rooms, remembering visited state |
| TileSorting-v0 |
Sort numbered tiles by pushing |
| PackingPuzzle-v0 |
Push objects onto matching target slots |
| PreciseNavigation-v0 |
Navigate ice terrain where agent slides; push boxes as stops |
| RecipeAssembly-v0 |
Collect ingredients in order and deliver to station |
| ToolUse-v0 |
Discover scroll combinations to create ORB and cross river |
| ResourceManagement-v0 |
Collect resources while managing energy drain |
Reasoning (8 tasks)
| Task |
Description |
| SwitchCircuit-v0 |
Activate color-coded switches to open barriers; cross-zone backtracking at hard+ |
| RuleInduction-v0 |
Discover hidden rules from ICE-marked cues, then INTERACT |
| LightsOut-v0 |
Turn all lights off (classic puzzle) |
| GraphColoring-v0 |
INTERACT to cycle node colors; no adjacent nodes share colors |
| SymbolMatching-v0 |
Match symbol pairs; matched pairs disappear |
| ProgramSynthesis-v0 |
Push gems onto reference pattern shown by scrolls |
| TaskInterference-v0 |
Balance GEM/ORB meters; collecting one drains the other |
| DeceptiveReward-v0 |
Reach true goal despite misleading reward signals |
Memory (4 tasks)
| Task |
Description |
| SequenceMemory-v0 |
Reproduce a sequence of colors (Simon Says) |
| DelayedGratification-v0 |
Resist immediate reward for larger delayed reward |
| TreasureHunt-v0 |
Find hidden treasures using directional scroll clues |
| FogOfWarExploration-v0 |
Navigate with persistent fog; only adjacent tiles visible |
Generalization (3 tasks)
| Task |
Description |
| FewShotAdaptation-v0 |
Watch demos to infer hidden rule, then navigate to correct target |
| DistributionShift-v0 |
Navigate maze with toggling walls and remapped actions |
| NoisyObservation-v0 |
Navigate with corrupted observations |
Multi-Agent (5 tasks)
| Task |
Description |
| CooperativeTransport-v0 |
Multiple agents push boxes to goal cooperatively |
| TagHunt-v0 |
Multi-agent chase/evade competition |
| ChaseEvade-v0 |
Evade Pacman-style enemies (Chaser, Ambusher, Patrol, Erratic) |
| Herding-v0 |
Herd fleeing sheep into a pen |
| EmergentStrategy-v0 |
Lure NPC sheep onto pressure plates; different NPC behaviors |
Difficulty Scaling
All tasks follow consistent scaling:
- Grid size: 5-8 (easy) → 20+ (expert)
- Complexity: More obstacles, constraints, and dependencies
- Episode length: Adjusted per difficulty