FBW-P01
Obstacle-Aware Navigation
Develop connectome-driven navigation that turns local sensory signals into steering and speed control in obstacle-filled virtual environments.
Training preparationApplications open
Statistics temporarily unavailable
Project background
Build a reproducible task for turning sensory signals into continuous action. A controlled two-dimensional environment will vary obstacle density, passages and sensory noise. The study asks whether neural responses support stable, interpretable steering in held-out scenes.
This project connects a defined neural-computation question to a reproducible evaluation protocol and a community participation pathway. Results will be interpreted only within their recorded model, task and evidence scope.
Training goals
- Increase goal arrival while reducing collisions and unproductive stops.
- Reduce excess travel distance and repeated left-right oscillation.
- Compare stability across starting points, maps and sensory noise.
Current foundation and milestones
The connectome model and response computation provide reusable foundations. A behavior-specific interface, continuous closed-loop executor, map benchmark and control results still require implementation and acceptance.
Next milestone: Freeze the 2D benchmark and sensory–action interface
- PlannedFreeze the 2D benchmark and sensory–action interface
- PlannedImplement the versioned task and measurement interface
- PlannedEvaluate frozen conditions and independent reproduction
Content updated:
Technical approach
The environment exposes left, front and right obstacle cues and relative goal direction. An encoder maps those signals to declared neural populations; a continuous runner reads activity from specified output populations and converts it to steering and speed. Initial searches cover input gain, temporal windows and constrained readout parameters. Arrival and collision measures are evaluated together so remaining motionless cannot count as success.
Method schematic
Task and evaluation design
Evaluation design
| Indicator | Definition | Direction | Units | Observation |
|---|---|---|---|---|
| Goal arrival | Fraction of fixed evaluation episodes that reach the goal. | Higher is better | % | Awaiting evaluation |
| Collision rate | Fraction of evaluation episodes with a defined obstacle contact. | Lower is better | % | Awaiting evaluation |
| Completion time | Model time to reach the declared goal region; failures and stall time are reported separately. | Lower is better | model time | Awaiting evaluation |
| Stall fraction | Fraction of the evaluation window spent below the protocol motion threshold. | Lower is better | % | Awaiting evaluation |
| Path length | Distance traveled before reaching the goal, compared under matched maps and starting positions. | Lower is better | environment distance | Awaiting evaluation |
No measured values have been published. Acceptance thresholds will be frozen in each task protocol before evaluation.
Expected deliverables
- A versioned navigation configuration package.
- Reproduction materials for fixed evaluation maps.
- A performance and failure-case report.
Ways to participate
- Parameter search
- Batch scene evaluation
- Independent reproduction
- Front-end environment development
Device requirements and task schedules will be matched by the community after runnable profiles have been evaluated. Unknown specifications are welcome in the application.
Apply to join training
Applications are received for community review and matching. Selection does not mean training has started.
References and related work
External sources provide technical background. They do not establish equivalent FBW capabilities.
View published research records ↗