Validation & Metrics

Benchmarks Inform Research, Pilots Prove Value

FieldSpace validates perception and plans with a physics-based world model, then prunes infeasible trajectories. In pilots we measure: false positive reduction, trajectory rejection rates for unsafe motion, decision latency, GPU hours per mile or per task, degraded comms task completion.

What We Measure in Pilots

These KPIs reflect our role as a physics layer beside your perception and control stack.

📉

False Positive Reduction

How many perception errors does FieldSpace catch? Measured as % reduction in unsafe detections that would have triggered bad plans.

Typical pilot results:

35-50%

⏱️

Decision Latency

Added latency from perception output to constrained trajectory. Critical for control loop timing. Measured in ms (p50, p95, p99).

Typical pilot results:

<5ms

💻

Compute Usage

GPU hours per mile or per task. FieldSpace prunes infeasible futures early, reducing need for heavy vision processing.

Typical pilot results:

20-30%

savings

🚫

Trajectory Rejection Rate

How often does FieldSpace reject unsafe motion before it reaches your controller? High rate = better safety envelope.

Typical pilot results:

15-25%

of candidate plans

📶

Degraded Comms Resilience

Task completion rate when network/perception bandwidth is reduced. FieldSpace continues locally with short-horizon prediction.

Typical pilot results:

95%+

safe operation maintained

🔧

Integration Time

Days from ROS/gRPC handshake to first on-vehicle or robot trial. Fast integration = faster time to value.

Typical pilot results:

5-10

days

How We Validate

Our pilot program measures FieldSpace impact at each stage, from log replay to on-vehicle trials.

1

Log Replay

Run FieldSpace on your recorded data. Baseline KPI measurement without on-vehicle risk.

Week 1 of pilot

2

Simulation

Optional: validate in your sim environment with edge case scenarios.

Week 1-2 (optional)

3

Closed Course

On-vehicle or robot trials with FieldSpace running in parallel to your stack. A/B testing with safety driver.

Week 2-3 of pilot

4

KPI Report

Comprehensive report on all KPIs with degraded-comms scenarios. Go/no-go decision.

Week 3 final deliverable

Research & Development Benchmarks

These academic benchmarks inform our R&D, but pilot KPIs prove value in your environment.

Internal Performance Metrics

Traffic Field Computation (PDE Solver)

Update Rate (GPU)1,974 FPS
Objects Tracked50+
Prediction Horizon0.5-3s

Perception Validation

Physics Constraint Check<1ms
Sensor Fusion (Kalman)<2ms
Trajectory Safety Check<3ms

Note: These are internal R&D benchmarks on reference hardware. Your pilot will measure real-world KPIs on your platform and data.

Validate FieldSpace on Your Data

Start a 4-week pilot. Week 1: log replay and baseline KPIs. Week 2-3: on-vehicle trials and A/B testing. Week 3: final report and go/no-go decision.

KPI-driven validation. Integrates with ROS 1/2, gRPC. Built for autonomy teams.