Summary of Key Points
Recently, the "haptic technology" segment within the field of embodied intelligence has suddenly become very popular: corporate orders have surged (Qianjue Robotics received several times more orders in the first half of the year than in the entire previous year), and there has been a rush of funding (at least 8 investments over 2.5 billion yuan since July, with two unicorns, Pashini and Yimu Technology, each raising 1 billion yuan). The reason for this surge is industry consensus that robots need to perform delicate tasks just like humans, which requires "haptic capabilities" (similar to how humans use their hands to sense force and position). Currently, there are two main technological approaches: hardware (installing sensors) and software (extracting haptic information from videos), but the industry still faces challenges such as limited data, difficulty in training, and high costs. Companies are working on finding solutions to overcome these obstacles.
Why Has the Haptic Technology Segment Suddenly Become So Popular?
The popularity of haptic technology is not accidental:
- Surging Orders: The founder of Qianjue Robotics stated that orders in the first half of this year were several times higher than in the entire previous year, indicating a sudden surge in market demand.
- Capital Inflow: There have been at least 8 investments over 2.5 billion yuan since July, with two unicorns (Pashini and Yimu Technology) each receiving 1 billion yuan, which demonstrates the significant interest from investors in this field.
- Industry Consensus: Experts agree that haptic capabilities are essential for robots to perform delicate tasks effectively; without them, robots would struggle with tasks like gently handling soft fruits or accurately placing headphones in charging cases.
Haptic Robots Can Perform Delicate Tasks That Were Previously Impossible
Haptic technology enhances a robot's "manual skills," as illustrated by the following examples:
- Gently Handling Soft Items: Robots equipped with haptic sensors can pick up chips and soft fruits without damaging them.
- Accurate Paper Folding: Even in the presence of interference, they can fold paper boxes correctly (whereas traditional robots fail due to visual obstruction).
- Precise Placement: Placing headphones in charging cases requires precise positioning (within a few millimeters). While vision may be obstructed or unable to detect slight misalignments, haptic sensors can adjust the grip accordingly.
- Grabbing Items from Packages: Traditional robots rely on cameras to view the contents of packages, but with haptic sensors, they can determine the position of items more accurately.
In simple terms, vision is like "eyes," while haptic technology provides the "sense of touch"—without it, robots are like blind hands that cannot perform tasks requiring fine-grained perception.
Two Main Technological Approaches
There are two primary methods for implementing haptic capabilities in robots:
1. Hardware-Based Approach: Installing "Haptic Gloves"
- Traditional Sensors: Such as piezoresistive and capacitive sensors (similar to those in smartphone touchscreens), but they are susceptible to electromagnetic interference and have limited accuracy.
- Advanced Vision-Haptic Sensors: Qianjue Robotics uses a soft gel material on the robot's grippers, which is photographed by cameras. Algorithms then analyze the deformation of the gel to determine the force and position applied. The advantages include high resolution and resistance to interference, as well as modular design (only the gel needs to be replaced, not the entire sensor).
2. Software-Based Approach: Learning from Human Videos
- Learning from Human Behavior: This approach directly extracts haptic information from videos of humans performing tasks. It uses physical principles and human experience to estimate contact forces and other parameters. The downside is lower accuracy compared to hardware-based methods, but it eliminates the need for additional hardware and allows for rapid data collection. However, high-quality hardware data is often combined to improve accuracy.
Challenges on the Road Ahead
Despite the popularity of haptic technology, there are several significant hurdles to overcome:
- Limited Data: High-quality haptic data is scarce, typically requiring tens of thousands of hours of training. The cost of sensors and equipment is also high, and factory data collection often lacks diversity (e.g., only focusing on specific tasks like picking up apples).
- Difficult Training: Haptic signals are only generated during contact, which makes it challenging to align them accurately with visual and motion data.
- High Costs: Sensor costs range from a few thousand to tens of thousands of yuan per unit. Additionally, model training can be extremely expensive, especially for explicit world models that require precise simulations of the physical world.
The Future of Haptic Technology
Despite these challenges, haptic technology is crucial for robots to perform practical tasks effectively. Companies are either developing foundational infrastructure (such as providing data and collection solutions) or collaborating with hardware manufacturers to provide advanced models (like ZhiZaiWuJie). As technology advances, haptic robots are expected to find widespread applications in manufacturing and service industries (e.g., parcel sorting, domestic services).
(End of Article)