The National Highway Traffic Safety Administration (NHTSA) has closed Preliminary Evaluation PE24033 regarding Tesla’s Actually Smart Summon (ASS) feature. The investigation, which focused on low-speed impacts during active sessions, concluded that incident rates were below 1% and typically resulted in minor property damage.
Highlights
- Six over-the-air (OTA) software updates were deployed to improve camera blockage detection and object perception.
- NHTSA found that incidents represented a fraction of 1% of millions of total Summon sessions.
- New high-fidelity occupancy networks were integrated to improve detection of dynamic objects like gate arms.
- The investigation noted that user error, including operating with obstructed camera views, contributed to most impacts.
Industry Context
The closure of this investigation highlights the evolving regulatory oversight of SAE Level 2 driver-assist systems. As manufacturers move toward vision-only architectures, the ability of these systems to detect environmental occlusions—such as snow or condensation—becomes a critical safety benchmark. Tesla’s use of iterative OTA updates to resolve “corner case” collisions, such as impacts with parking bollards and gate arms, reflects the current industry trend of real-time safety refinement through software.
Technical Improvements and Perception Updates
During the evaluation period, the Office of Defects Investigations (ODI) identified specific failure modes related to camera blockages. In instances involving snowy conditions, the system failed to detect obstructed forward-facing cameras, leading to collisions with unoccupied vehicles.
Tesla responded with a series of firmware updates to mitigate these risks:
- Camera Blockage Detection: OTA updates 578998 and 579185 improved mechanisms to identify when sensors are obscured.
- Visibility Checks: Updates SW-578752 and SW-580322 reduced false negatives caused by snow or condensation.
- Neural Network Integration: On November 20, 2025, Tesla deployed SW-580514, adding object detections from a secondary neural network to enhance spatial reconstruction.
Occupancy Network Enhancements
A specific failure mode involved the system’s inability to yield for gate arms in garage exit lanes. To address this, Tesla deployed SW-578839 on February 6, 2025. This update introduced a high-fidelity occupancy determination network.
This network utilizes raw data from vehicle sensory systems to reconstruct field objects with higher accuracy. The ODI noted that while the system has been improved, app users often failed to command a stop despite having access to the camera stream. The agency stated that the closing of the investigation does not preclude future actions if new safety defects emerge.
Sign up for our weekly emails to stay up to date on the latest news!
