Outrider deploys advanced reinforcement learning models for enhanced autonomous yard ops

Outrider, a provider of autonomous yard operations for logistics hubs, announced what it says is the industry’s first deployment of advanced reinforcement learning (RL) techniques to maximize freight throughput at customer sites.

Outrider’s RL models, according to the company, increase path planning speed by 10x and enable the Outrider System to move freight more efficiently and safely through busy, complex distribution yards.

AI-driven redundant safety mechanisms

Outrider’s AI-driven capabilities are complemented by what the company said are industry-first redundant safety mechanisms, merging the benefits of AI with traditional functional safety approaches used for industrial operations. Outrider has addressed over 200,000 safety scenarios, and multiple third-party safety experts and Fortune 500 customers have validated its safety case.

“Using the latest advances in AI, Outrider is continually decreasing the turn time of trailers moved autonomously in logistics yards,” said Vittorio Ziparo, CTO and executive vice president of engineering at Outrider. “By training and evaluating our system performance with RL in simulation and real-world scenarios, our customers see incremental improvements in speed and efficiency with our technology.”

RL techniques involve creating a model that improves decision-making over time. Using years of data samples of behaviors, Outrider developed an RL curriculum with increasing difficulty so the model can learn. This technique reinforces preferred behaviors, such as following traffic rules and maintaining safe distances from other vehicles, and discourages undesirable behaviors. Once the RL models are tested extensively in simulation and on-vehicle at Outrider’s Advanced Testing Facility, the model and code are deployed into autonomous operations at customer sites.

“Our Fortune 500 customers’ yards are complex, with hundreds of trucks, trailers, other vehicles and pedestrians operating onsite daily,” Ziparo said. “RL is critical to automating these yards at scale because it enables our commercial system to handle increasingly complex and diverse environments – from distribution and manufacturing yards to intermodal and port terminals.”

Outrider utilizes reinforcement learning, deep learning

Outrider’s RL techniques use millions of proprietary, yard-specific data points collected and labeled across various large, complex distribution yards in multiple industries. These data points feed Outrider’s proprietary deep learning (DL) and RL models to create neural networks that automate yard tasks with increasing intelligence, precision and speed.

Outrider’s private AI cloud deployment utilizes NVIDIA’s DGX H200 GPUs installed at a Denver-based data center owned and operated by Equinix.

“When dealing with exponentially increasing amounts of data to train DL and RL models, processing speed and training velocity per dollar spent matters,” said Tom Baroch, senior director of global partnerships at Outrider. “NVIDIA, an investor in Outrider, helped us secure the cutting-edge hardware necessary to double our DL training speed and we deployed the hybrid cloud training environment, which increased training velocity per dollar by six times. Taking this approach, Outrider delivers even greater value sooner to our customers.”

Using RL further complements Outrider’s other offerings, including trailer moves, hitching, backing, trailer brake line connection, yard inventory tracking, and integration with warehouse, yard and transportation management systems.