Real-Time High-μ Friction Estimation for L3+ Autonomous EVs – How Easyrain DAI Enables Safe Power Management
High-torque electric vehicles operating under Level 3 autonomous driving conditions require the vehicle control system to make continuous, high-confidence decisions about the tire-road interface — without a driver as a fallback. Easyrain’s DAI Virtual Sensor Platform addresses this requirement by estimating the tire-road friction coefficient (μ) in real time using only pre-existing onboard vehicle data, enabling L3+ systems to safely and dynamically manage motor torque output.
What Happened: The Technical Facts
Easyrain DAI processes wheel speed, yaw rate, and steering angle data — signals already present in the vehicle’s existing sensor network — to compute a continuous estimate of the tire-road friction coefficient (μ). No additional hardware is required.
The system applies a hybrid algorithmic architecture including a Kalman Filter for real-time state estimation with a CNN vision-based model for surface-type classification. A layer then adapts the output to the specific vehicle’s dynamic signature, compensating for differences in axle load, tire compound, and suspension tuning across OEM platforms.
The key output is a binary-class contextual signal: high-μ (μ > 0.5) versus low-μ (wet, gravel, snow, or unknown surface < 0.5). This classification is computed and delivered to the ECU with a latency of under 150 milliseconds, meeting the actuation budget required by L3 functional safety architectures.
Why It Matters: Torque Management and Autonomous Control
Chinese EV OEMs developing L3-ready platforms need to know with precision when the tire-road friction coefficient is at its optimum (μ ≈ 1 — confirmed dry asphalt) in order to safely authorize maximum motor performance for the driver or the autonomous planner. Conversely, when road conditions are sub-optimal — rain, wet asphalt, loose gravel, snow — manufacturers apply an internal power ceiling to prevent uncontrolled wheel slip. That ceiling is not a fixed universal value: each OEM defines its own threshold, based on vehicle architecture, tire compound, and safety policy.
What DAI enables — through the same algorithms already used to estimate μ in real time — is the ability to detect conditions of sustained lateral acceleration (constant IMU) that may call for a power limitation, and to suggest the power level compatible with maximum safety given current road conditions. This is not a fixed percentage enforced by a market-standard system: it is a prospective capability that Easyrain DAI brings to OEM platforms as a software-defined sensor input, without any additional hardware.
This matters for two distinct reasons:
- Performance and handling: High-torque EVs — particularly dual- and quad-motor platforms common among Chinese OEMs — are most effective when traction limits are known with precision. Torque vectoring algorithms depend directly on accurate μ estimates to distribute power between axles without initiating understeer or oversteer events.
- Autonomous planner coherence: An L3 path planner generates trajectory commands assuming a defined friction envelope. If the μ input is absent or estimated conservatively, the system either under-performs or applies unnecessary safety margins that degrade passenger comfort and range efficiency.
Key Data
- μ classification threshold: > 0.9 identifies confirmed dry asphalt (high-μ state)
- Estimation error: < 35%, compliant with ISO 26262 functional safety requirements for ADAS sensor inputs
- Signal latency: < 50 ms from sensor input to ECU output — within the ≤ 100 ms end-to-end actuation budget documented for L3 urban and highway systems
- Power suggestion: DAI signals the power level compatible with maximum safety based on current road conditions — the OEM defines the actual actuation threshold internally
- Hardware addition: zero — operates exclusively on signals already present in the vehicle’s CAN/Ethernet bus
- Deployment architecture: edge-only, no cloud dependency, V2X-ready for cooperative road-state sharing
Market and Regulatory Context
China’s MIIT granted its first product market access approvals for Level 3 conditional autonomous driving in December 2025, with compliant models including the Changan Deepal SL03 and BAIC Arcfox αS6, authorizing hands-free L3 operation on designated highways and urban expressways.
BYD initiated large-scale L3 internal testing in Shenzhen in late 2025, accumulating over 150,000 km of real-world data under complex conditions. XPeng secured official L3 road test permits in Guangzhou, and Xiaomi Automobile received an L3 road test license in Beijing during the same period.
The broader market context reinforces the scale of the opportunity: New Energy Vehicle (NEV) sales in China reached 16.49 million units in 2025, accounting for 47.9% of total new vehicle sales — a market in which L3-ready EV platforms are rapidly becoming a competitive baseline.
Under ISO 26262 ASIL D, systems governing vehicle dynamics on L3+ platforms — including torque management and braking actuation — must meet the most stringent safety integrity classification. DAI’s < 150 ms latency are designed to operate within these constraints, allowing OEMs to integrate the μ signal into safety-critical torque-control loops without additional validation overhead for the sensor hardware layer.
China’s regulatory preference for edge-only processing (without cloud-dependent inference) and the national push for V2X integration under the C-V2X standard make DAI’s architecture — local compute, no external data dependency, V2X-compatible output — directly compatible with compliance requirements for domestic L3 homologation.
What to Expect
The global Level 3 vehicle market is projected to grow from approximately 291,000 units in 2025 to 8.7 million units by 2035, at a 40.5% CAGR, with Asia-Pacific as the primary growth region. As more Chinese OEMs submit L3 models for MIIT market access approval, the need for ISO 26262-compliant, hardware-free sensor inputs — particularly for dynamic surfaces — will intensify.
Easyrain’s modular ecosystem extends beyond friction estimation. The AIS Active Safety System provides a physical grip-restoration layer on wet surfaces, validated at −20% braking distance and +225% lateral traction improvement in aquaplaning conditions. The ERC Cloud Platform enables fleet-level aggregation of road-state data, supporting predictive maintenance and cooperative hazard mapping — functions that V2X networks are designed to distribute at scale.
For Chinese EV OEMs building L3-ready platforms, the DAI signal represents a software-defined sensor input that can be mapped to ECU torque logic without changes to the vehicle’s physical hardware configuration, aligning with the software-defined vehicle (SDV) architecture model that domestic OEMs are actively adopting.
Frequently Asked Questions
Q: How does Easyrain DAI estimate tire-road friction without additional sensors?
A: Easyrain DAI uses signals already present in the vehicle’s onboard network — wheel speed, yaw rate, and steering angle data — and processes them through a hybrid algorithm combining including a Kalman Filter, a CNN-based surface classifier, and a Dual Extended Unscented Kalman Filter (DeUKF). The DeUKF adapts the estimation model to the specific vehicle’s dynamic characteristics, producing a friction coefficient (μ) estimate with less than 35% error and no requirement for additional hardware.
Q: How does DAI’s friction signal help OEMs manage motor power safely on L3 autonomous EVs?
A: Chinese EV OEMs need to know when road friction is at its optimum (μ ≈ 1) to safely authorize maximum motor performance, and when conditions are sub-optimal — rain, wet asphalt, gravel — to apply a power ceiling that prevents uncontrolled wheel slip. Each manufacturer defines that ceiling internally, based on vehicle architecture and safety policy. DAI provides the real-time μ estimate that makes this decision possible: using the same algorithms that classify the road surface, it can also detect conditions requiring a power limitation and suggest the power level compatible with maximum safety given current road conditions. This capability is not a fixed market standard — it is a software-defined value that Easyrain DAI brings to L3 platforms without any additional hardware.
Q: Why is DAI’s edge-only architecture relevant to Chinese L3 OEM requirements?
A: Chinese regulatory frameworks for L3 autonomous driving place restrictions on cloud-dependent safety-critical functions, requiring that core vehicle control decisions be executed locally on onboard compute hardware. DAI operates entirely at the edge — performing all friction estimation on the vehicle’s existing processing units without external data queries or cloud inference. Its architecture is also compatible with C-V2X output, allowing road-state data to be shared cooperatively with infrastructure and other vehicles. This combination of local processing and V2X integration aligns with China’s domestic L3 homologation requirements and the software-defined vehicle (SDV) model adopted by major Chinese EV manufacturers.
